{ "schema_version": 4, "approaches": [ { "id": "airbnb-airchat", "company": "Airbnb", "agent_name": "Airchat (airchat-cli)", "approach_type": "platform", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025", "source_id": "airbnb-airchat-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "coding task → reviewed pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "interactive", "background" ], "state": "unknown", "identity": "unknown", "evidence_strength": "mixed" }, "claim_ids": [ "airbnb-airchat--summary", "airbnb-airchat--headline-metric", "airbnb-airchat--architecture-harness", "airbnb-airchat--architecture-model", "airbnb-airchat--architecture-tool-access", "airbnb-airchat--key-metrics-0", "airbnb-airchat--operating-models-0" ], "source_ids": [ "airbnb-airchat-source-1", "airbnb-airchat-source-2", "airbnb-airchat-source-3" ], "interfaces": [] }, { "id": "atlassian-rovo-dev", "company": "Atlassian", "agent_name": "Rovo Dev (RovoDev)", "aliases": [ "RovoDev", "HULA" ], "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2024, "first_public_evidence": { "date": "2024-11", "source_id": "atlassian-rovo-dev-source-3" }, "last_reviewed_at": "2026-09-09", "status": "commercialized", "domains": [ "coding", "code-review" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "Jira issue → reviewed pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "atlassian-rovo-dev--summary", "atlassian-rovo-dev--headline-metric", "atlassian-rovo-dev--architecture-harness", "atlassian-rovo-dev--architecture-interfaces", "atlassian-rovo-dev--key-metrics-0", "atlassian-rovo-dev--key-metrics-1", "atlassian-rovo-dev--operating-models-0" ], "source_ids": [ "atlassian-rovo-dev-source-1", "atlassian-rovo-dev-source-2", "atlassian-rovo-dev-source-3" ], "interfaces": [ "jira" ] }, { "id": "block-builderbot", "company": "Block", "agent_name": "Builderbot", "approach_type": "orchestration-system", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "block-builderbot-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "ticket → reviewed pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "block-builderbot--summary", "block-builderbot--headline-metric", "block-builderbot--architecture-sandbox", "block-builderbot--architecture-harness", "block-builderbot--architecture-model", "block-builderbot--architecture-interfaces", "block-builderbot--architecture-tool-access", "block-builderbot--architecture-knowledge", "block-builderbot--primitives-0", "block-builderbot--primitives-1", "block-builderbot--key-metrics-0", "block-builderbot--key-metrics-1", "block-builderbot--key-metrics-2", "block-builderbot--lessons-learned-0", "block-builderbot--lessons-learned-1", "block-builderbot--lessons-learned-2", "block-builderbot--operating-models-0" ], "source_ids": [ "block-builderbot-source-1", "block-builderbot-source-2", "block-builderbot-source-3", "block-builderbot-source-4", "block-builderbot-source-5" ], "interfaces": [ "slack", "linear", "jira", "github" ] }, { "id": "brex-agent-platform", "company": "Brex", "agent_name": "Internal Agent Platform", "approach_type": "platform", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025-09-25", "source_id": "brex-agent-platform-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "finance-ops", "support", "customer-success" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "internal operations request → completed operation", "attention_boundary": "continuous-steering", "level": 2 } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "secondary-only" }, "claim_ids": [ "brex-agent-platform--summary", "brex-agent-platform--headline-metric", "brex-agent-platform--architecture-sandbox", "brex-agent-platform--architecture-harness", "brex-agent-platform--architecture-model", "brex-agent-platform--architecture-interfaces", "brex-agent-platform--architecture-tool-access", "brex-agent-platform--architecture-knowledge", "brex-agent-platform--architecture-credentials", "brex-agent-platform--primitives-0", "brex-agent-platform--primitives-1", "brex-agent-platform--key-metrics-0", "brex-agent-platform--key-metrics-1", "brex-agent-platform--key-metrics-2", "brex-agent-platform--key-metrics-3", "brex-agent-platform--lessons-learned-0", "brex-agent-platform--lessons-learned-1", "brex-agent-platform--lessons-learned-2", "brex-agent-platform--lessons-learned-3", "brex-agent-platform--lessons-learned-4", "brex-agent-platform--operating-models-0" ], "source_ids": [ "brex-agent-platform-source-1" ], "interfaces": [ "slack", "internal-ui" ] }, { "id": "browserbase-bb", "company": "Browserbase", "agent_name": "bb", "approach_type": "task-agent", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "browserbase-bb-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review", "support", "customer-success", "research" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "coding request → reviewed pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "browserbase-bb--summary", "browserbase-bb--headline-metric", "browserbase-bb--architecture-sandbox", "browserbase-bb--architecture-harness", "browserbase-bb--architecture-model", "browserbase-bb--architecture-interfaces", "browserbase-bb--architecture-tool-access", "browserbase-bb--architecture-knowledge", "browserbase-bb--architecture-credentials", "browserbase-bb--primitives-0", "browserbase-bb--primitives-1", "browserbase-bb--primitives-2", "browserbase-bb--key-metrics-0", "browserbase-bb--key-metrics-1", "browserbase-bb--key-metrics-2", "browserbase-bb--lessons-learned-0", "browserbase-bb--lessons-learned-1", "browserbase-bb--lessons-learned-2", "browserbase-bb--lessons-learned-3", "browserbase-bb--operating-models-0" ], "source_ids": [ "browserbase-bb-source-1" ], "interfaces": [ "slack", "web", "webhook" ] }, { "id": "cloudflare-ai-stack", "company": "Cloudflare", "agent_name": "Internal AI engineering stack", "approach_type": "platform", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "cloudflare-ai-stack-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "pull request → AI review findings", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "cloudflare-ai-stack--summary", "cloudflare-ai-stack--headline-metric", "cloudflare-ai-stack--architecture-sandbox", "cloudflare-ai-stack--architecture-harness", "cloudflare-ai-stack--architecture-model", "cloudflare-ai-stack--architecture-interfaces", "cloudflare-ai-stack--architecture-tool-access", "cloudflare-ai-stack--architecture-knowledge", "cloudflare-ai-stack--architecture-credentials", "cloudflare-ai-stack--architecture-context-mgmt", "cloudflare-ai-stack--primitives-0", "cloudflare-ai-stack--primitives-1", "cloudflare-ai-stack--primitives-2", "cloudflare-ai-stack--primitives-3", "cloudflare-ai-stack--key-metrics-0", "cloudflare-ai-stack--key-metrics-1", "cloudflare-ai-stack--key-metrics-2", "cloudflare-ai-stack--key-metrics-3", "cloudflare-ai-stack--lessons-learned-0", "cloudflare-ai-stack--lessons-learned-1", "cloudflare-ai-stack--lessons-learned-2", "cloudflare-ai-stack--lessons-learned-3", "cloudflare-ai-stack--operating-models-0" ], "source_ids": [ "cloudflare-ai-stack-source-1", "cloudflare-ai-stack-source-2" ], "interfaces": [ "cli", "ci", "web" ] }, { "id": "coinbase-forge-mux", "company": "Coinbase", "agent_name": "Forge / Mux", "approach_type": "agent-system", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "coinbase-forge-mux-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "Slack, GitHub, or Linear request → reviewed pull request and build", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "coinbase-forge-mux--summary", "coinbase-forge-mux--headline-metric", "coinbase-forge-mux--architecture-sandbox", "coinbase-forge-mux--architecture-harness", "coinbase-forge-mux--architecture-model", "coinbase-forge-mux--architecture-interfaces", "coinbase-forge-mux--architecture-tool-access", "coinbase-forge-mux--architecture-knowledge", "coinbase-forge-mux--primitives-0", "coinbase-forge-mux--primitives-1", "coinbase-forge-mux--key-metrics-0", "coinbase-forge-mux--lessons-learned-0", "coinbase-forge-mux--lessons-learned-1", "coinbase-forge-mux--lessons-learned-2", "coinbase-forge-mux--operating-models-0" ], "source_ids": [ "coinbase-forge-mux-source-1", "coinbase-forge-mux-source-2" ], "interfaces": [ "slack", "github", "linear" ] }, { "id": "databricks-costar", "company": "Databricks", "agent_name": "coSTAR and internal engineering agents", "approach_type": "agent-system", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025", "source_id": "databricks-costar-source-1" }, "last_reviewed_at": "2026-08-31", "status": "internal", "domains": [ "coding", "code-review", "on-call" ], "autonomy": "unknown", "operating_models": [ { "scope": "internal engineering workflows → agent-produced changes", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "interactive", "background" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "databricks-costar--summary", "databricks-costar--architecture-harness", "databricks-costar--architecture-knowledge", "databricks-costar--key-metrics-0", "databricks-costar--key-metrics-1", "databricks-costar--operating-models-0" ], "source_ids": [ "databricks-costar-source-1", "databricks-costar-source-2" ], "interfaces": [] }, { "id": "domu-clementino", "company": "Domu", "agent_name": "Clementino", "approach_type": "task-agent", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "domu-clementino-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "support", "finance-ops", "coding", "recruitment", "customer-success" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "employee request → approved customer-impacting action", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "domu-clementino--summary", "domu-clementino--headline-metric", "domu-clementino--architecture-harness", "domu-clementino--architecture-model", "domu-clementino--architecture-interfaces", "domu-clementino--architecture-tool-access", "domu-clementino--architecture-knowledge", "domu-clementino--architecture-credentials", "domu-clementino--architecture-context-mgmt", "domu-clementino--primitives-0", "domu-clementino--primitives-1", "domu-clementino--key-metrics-0", "domu-clementino--lessons-learned-0", "domu-clementino--lessons-learned-1", "domu-clementino--lessons-learned-2", "domu-clementino--operating-models-0" ], "source_ids": [ "domu-clementino-source-1" ], "interfaces": [ "slack", "desktop" ] }, { "id": "doordash-code-review", "company": "DoorDash", "agent_name": "AI Code Review Agent", "relationships": [ { "type": "component-of", "approach_id": "doordash-flux" } ], "approach_type": "background-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "doordash-code-review-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "pull request → AI review comments", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "background", "event-driven" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "doordash-code-review--summary", "doordash-code-review--headline-metric", "doordash-code-review--architecture-harness", "doordash-code-review--architecture-model", "doordash-code-review--architecture-interfaces", "doordash-code-review--architecture-tool-access", "doordash-code-review--architecture-knowledge", "doordash-code-review--primitives-0", "doordash-code-review--key-metrics-0", "doordash-code-review--key-metrics-1", "doordash-code-review--lessons-learned-0", "doordash-code-review--lessons-learned-1", "doordash-code-review--operating-models-0" ], "source_ids": [ "doordash-code-review-source-1" ], "interfaces": [ "github" ] }, { "id": "doordash-flux", "company": "DoorDash", "agent_name": "Flux / Agentic AI Platform", "approach_type": "platform", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025-11-11", "source_id": "doordash-flux-source-2" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "code-review", "coding", "ci-triage", "on-call", "maintenance", "data" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "engineering task → reviewed agent output", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive", "scheduled" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "doordash-flux--summary", "doordash-flux--headline-metric", "doordash-flux--architecture-sandbox", "doordash-flux--architecture-harness", "doordash-flux--architecture-model", "doordash-flux--architecture-interfaces", "doordash-flux--architecture-tool-access", "doordash-flux--architecture-knowledge", "doordash-flux--architecture-credentials", "doordash-flux--architecture-context-mgmt", "doordash-flux--primitives-0", "doordash-flux--primitives-1", "doordash-flux--primitives-2", "doordash-flux--primitives-3", "doordash-flux--primitives-4", "doordash-flux--key-metrics-0", "doordash-flux--key-metrics-1", "doordash-flux--key-metrics-2", "doordash-flux--lessons-learned-0", "doordash-flux--lessons-learned-1", "doordash-flux--lessons-learned-2", "doordash-flux--lessons-learned-3", "doordash-flux--lessons-learned-4", "doordash-flux--lessons-learned-5", "doordash-flux--operating-models-0" ], "source_ids": [ "doordash-flux-source-1", "doordash-flux-source-2", "doordash-flux-source-3" ], "interfaces": [ "slack", "github", "scheduled", "cli", "skill", "cursor" ] }, { "id": "dropbox-nova", "company": "Dropbox", "agent_name": "Nova", "approach_type": "platform", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "dropbox-nova-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "ci-triage", "on-call", "maintenance" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "agent-assisted SDLC workflow → accepted change", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "dropbox-nova--summary", "dropbox-nova--headline-metric", "dropbox-nova--architecture-sandbox", "dropbox-nova--architecture-harness", "dropbox-nova--architecture-model", "dropbox-nova--architecture-interfaces", "dropbox-nova--architecture-tool-access", "dropbox-nova--architecture-knowledge", "dropbox-nova--architecture-credentials", "dropbox-nova--architecture-context-mgmt", "dropbox-nova--primitives-0", "dropbox-nova--primitives-1", "dropbox-nova--key-metrics-0", "dropbox-nova--key-metrics-1", "dropbox-nova--key-metrics-2", "dropbox-nova--lessons-learned-0", "dropbox-nova--lessons-learned-1", "dropbox-nova--lessons-learned-2", "dropbox-nova--lessons-learned-3", "dropbox-nova--operating-models-0" ], "source_ids": [ "dropbox-nova-source-1", "dropbox-nova-source-2" ], "interfaces": [ "web", "cli", "api", "slack" ] }, { "id": "flex-investigation-agent", "company": "Flex", "agent_name": "AI Investigation Agent", "approach_type": "task-agent", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "flex-investigation-agent-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "finance-ops", "on-call", "coding" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "payment investigation → proposed code fix", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "flex-investigation-agent--summary", "flex-investigation-agent--architecture-sandbox", "flex-investigation-agent--architecture-harness", "flex-investigation-agent--architecture-model", "flex-investigation-agent--architecture-interfaces", "flex-investigation-agent--architecture-tool-access", "flex-investigation-agent--primitives-0", "flex-investigation-agent--lessons-learned-0", "flex-investigation-agent--lessons-learned-1", "flex-investigation-agent--operating-models-0" ], "source_ids": [ "flex-investigation-agent-source-1" ], "interfaces": [ "slack" ] }, { "id": "github-qubot", "company": "GitHub", "agent_name": "Qubot", "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026-06", "source_id": "github-qubot-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "data" ], "autonomy": "assistive", "operating_models": [ { "scope": "data question → warehouse answer", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "github-qubot--summary", "github-qubot--headline-metric", "github-qubot--architecture-model", "github-qubot--architecture-tool-access", "github-qubot--key-metrics-0", "github-qubot--key-metrics-1", "github-qubot--operating-models-0" ], "source_ids": [ "github-qubot-source-1" ], "interfaces": [] }, { "id": "harvey-spectre", "company": "Harvey", "agent_name": "Spectre", "approach_type": "platform", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "harvey-spectre-source-1" }, "last_reviewed_at": "2026-08-31", "status": "internal", "domains": [ "coding", "code-review", "on-call", "security" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "incident or request → reviewable diff or pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "harvey-spectre--summary", "harvey-spectre--architecture-sandbox", "harvey-spectre--architecture-harness", "harvey-spectre--architecture-model", "harvey-spectre--architecture-interfaces", "harvey-spectre--architecture-tool-access", "harvey-spectre--architecture-credentials", "harvey-spectre--architecture-context-mgmt", "harvey-spectre--primitives-0", "harvey-spectre--primitives-1", "harvey-spectre--lessons-learned-0", "harvey-spectre--lessons-learned-1", "harvey-spectre--lessons-learned-2", "harvey-spectre--operating-models-0" ], "source_ids": [ "harvey-spectre-source-1", "harvey-spectre-source-2" ], "interfaces": [ "slack", "web", "automation" ] }, { "id": "hubspot-sidekick", "company": "HubSpot", "agent_name": "Sidekick", "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026-03", "source_id": "hubspot-sidekick-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "pull request → AI review comments", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "hubspot-sidekick--summary", "hubspot-sidekick--headline-metric", "hubspot-sidekick--architecture-harness", "hubspot-sidekick--architecture-sandbox", "hubspot-sidekick--architecture-tool-access", "hubspot-sidekick--architecture-interfaces", "hubspot-sidekick--key-metrics-0", "hubspot-sidekick--key-metrics-1", "hubspot-sidekick--key-metrics-2", "hubspot-sidekick--operating-models-0" ], "source_ids": [ "hubspot-sidekick-source-1", "hubspot-sidekick-source-2" ], "interfaces": [ "github" ] }, { "id": "linear-agent", "company": "Linear", "agent_name": "Linear Agent", "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "linear-agent-source-1" }, "last_reviewed_at": "2026-09-09", "status": "commercialized", "domains": [ "support", "customer-success", "coding" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "assigned coding work → agent-created change", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "linear-agent--summary", "linear-agent--architecture-sandbox", "linear-agent--architecture-harness", "linear-agent--architecture-model", "linear-agent--architecture-interfaces", "linear-agent--architecture-tool-access", "linear-agent--architecture-knowledge", "linear-agent--architecture-credentials", "linear-agent--primitives-0", "linear-agent--primitives-1", "linear-agent--primitives-2", "linear-agent--lessons-learned-0", "linear-agent--lessons-learned-1", "linear-agent--lessons-learned-2", "linear-agent--lessons-learned-3", "linear-agent--lessons-learned-4", "linear-agent--operating-models-0" ], "source_ids": [ "linear-agent-source-1", "linear-agent-source-2", "linear-agent-source-3", "linear-agent-source-4", "linear-agent-source-5", "linear-agent-source-6" ], "interfaces": [ "slack", "intercom", "linear", "github" ] }, { "id": "microsoft-prassistant", "company": "Microsoft", "agent_name": "PRAssistant", "approach_type": "background-agent", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025-07", "source_id": "microsoft-prassistant-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "pull request → AI review comments", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "microsoft-prassistant--summary", "microsoft-prassistant--headline-metric", "microsoft-prassistant--architecture-interfaces", "microsoft-prassistant--key-metrics-0", "microsoft-prassistant--key-metrics-1", "microsoft-prassistant--key-metrics-2", "microsoft-prassistant--operating-models-0" ], "source_ids": [ "microsoft-prassistant-source-1" ], "interfaces": [ "github" ] }, { "id": "monday-sphera-atlas-morphex", "company": "monday.com", "agent_name": "Sphera / Atlas / Morphex", "approach_type": "agent-system", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "monday-sphera-atlas-morphex-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review" ], "autonomy": "autonomous", "operating_models": [ { "scope": "Atlas or Morphex feature task → tested and merged pull request", "attention_boundary": "outcome-review", "level": 4 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "dedicated-agent", "evidence_strength": "secondary-only" }, "claim_ids": [ "monday-sphera-atlas-morphex--summary", "monday-sphera-atlas-morphex--headline-metric", "monday-sphera-atlas-morphex--architecture-sandbox", "monday-sphera-atlas-morphex--architecture-harness", "monday-sphera-atlas-morphex--architecture-model", "monday-sphera-atlas-morphex--architecture-interfaces", "monday-sphera-atlas-morphex--architecture-tool-access", "monday-sphera-atlas-morphex--architecture-knowledge", "monday-sphera-atlas-morphex--architecture-credentials", "monday-sphera-atlas-morphex--architecture-context-mgmt", "monday-sphera-atlas-morphex--primitives-0", "monday-sphera-atlas-morphex--primitives-1", "monday-sphera-atlas-morphex--primitives-2", "monday-sphera-atlas-morphex--primitives-3", "monday-sphera-atlas-morphex--key-metrics-0", "monday-sphera-atlas-morphex--key-metrics-1", "monday-sphera-atlas-morphex--key-metrics-2", "monday-sphera-atlas-morphex--key-metrics-3", "monday-sphera-atlas-morphex--lessons-learned-0", "monday-sphera-atlas-morphex--lessons-learned-1", "monday-sphera-atlas-morphex--lessons-learned-2", "monday-sphera-atlas-morphex--lessons-learned-3", "monday-sphera-atlas-morphex--lessons-learned-4", "monday-sphera-atlas-morphex--operating-models-0" ], "source_ids": [ "monday-sphera-atlas-morphex-source-1" ], "interfaces": [ "slack", "monday", "github" ] }, { "id": "notion-custom-agents", "company": "Notion", "agent_name": "Custom Agents", "approach_type": "platform", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026-04", "source_id": "notion-custom-agents-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "support", "finance-ops", "recruitment", "security" ], "autonomy": "unknown", "operating_models": [ { "scope": "cross-team internal tasks → Custom Agents output", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "notion-custom-agents--summary", "notion-custom-agents--headline-metric", "notion-custom-agents--key-metrics-0", "notion-custom-agents--key-metrics-1", "notion-custom-agents--key-metrics-2", "notion-custom-agents--operating-models-0" ], "source_ids": [ "notion-custom-agents-source-1", "notion-custom-agents-source-2" ], "interfaces": [] }, { "id": "plaid-ai-annotator", "company": "Plaid", "agent_name": "AI Annotator", "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025-06", "source_id": "plaid-ai-annotator-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "data" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "raw transactions → labeled training data", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "background" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "plaid-ai-annotator--summary", "plaid-ai-annotator--headline-metric", "plaid-ai-annotator--architecture-knowledge", "plaid-ai-annotator--key-metrics-0", "plaid-ai-annotator--key-metrics-1", "plaid-ai-annotator--operating-models-0" ], "source_ids": [ "plaid-ai-annotator-source-1" ], "interfaces": [] }, { "id": "plaid-fix-my-connection", "company": "Plaid", "agent_name": "Fix My Connection", "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025-06", "source_id": "plaid-fix-my-connection-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "ops", "maintenance" ], "autonomy": "autonomous", "operating_models": [ { "scope": "integration degradation → repaired connection", "attention_boundary": "outcome-review", "level": 4 } ], "rubric": { "invocation": [ "background" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "plaid-fix-my-connection--summary", "plaid-fix-my-connection--headline-metric", "plaid-fix-my-connection--architecture-tool-access", "plaid-fix-my-connection--key-metrics-0", "plaid-fix-my-connection--key-metrics-1", "plaid-fix-my-connection--operating-models-0" ], "source_ids": [ "plaid-fix-my-connection-source-1" ], "interfaces": [] }, { "id": "plaid-internal-mcp-server", "company": "Plaid", "agent_name": "Internal MCP server", "approach_type": "supporting-pattern", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025", "source_id": "plaid-internal-mcp-server-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding" ], "autonomy": "unknown", "operating_models": [ { "scope": "engineer request → internal tool access", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "plaid-internal-mcp-server--summary", "plaid-internal-mcp-server--headline-metric", "plaid-internal-mcp-server--architecture-harness", "plaid-internal-mcp-server--architecture-tool-access", "plaid-internal-mcp-server--architecture-credentials", "plaid-internal-mcp-server--key-metrics-0", "plaid-internal-mcp-server--key-metrics-1", "plaid-internal-mcp-server--operating-models-0" ], "source_ids": [ "plaid-internal-mcp-server-source-1" ], "interfaces": [] }, { "id": "posthog-stamphog", "company": "PostHog", "agent_name": "StampHog", "approach_type": "background-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026-07-09", "source_id": "posthog-stamphog-source-1" }, "last_reviewed_at": "2026-09-09", "status": "open-sourced", "domains": [ "code-review" ], "autonomy": "autonomous", "operating_models": [ { "scope": "eligible pull request → approval decision", "attention_boundary": "exception-only", "level": 5 } ], "rubric": { "invocation": [ "background", "event-driven" ], "state": "run-only", "identity": "dedicated-agent", "evidence_strength": "detailed-primary" }, "claim_ids": [ "posthog-stamphog--summary", "posthog-stamphog--headline-metric", "posthog-stamphog--architecture-harness", "posthog-stamphog--architecture-model", "posthog-stamphog--architecture-interfaces", "posthog-stamphog--architecture-tool-access", "posthog-stamphog--architecture-knowledge", "posthog-stamphog--architecture-credentials", "posthog-stamphog--architecture-context-mgmt", "posthog-stamphog--primitives-0", "posthog-stamphog--primitives-1", "posthog-stamphog--primitives-2", "posthog-stamphog--key-metrics-0", "posthog-stamphog--key-metrics-1", "posthog-stamphog--key-metrics-2", "posthog-stamphog--lessons-learned-0", "posthog-stamphog--lessons-learned-1", "posthog-stamphog--lessons-learned-2", "posthog-stamphog--operating-models-0" ], "source_ids": [ "posthog-stamphog-source-1", "posthog-stamphog-source-2", "posthog-stamphog-source-3" ], "interfaces": [ "github", "ci" ] }, { "id": "ramp-inspect", "company": "Ramp", "agent_name": "Inspect", "approach_type": "background-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "ramp-inspect-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review", "on-call" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "Inspect coding task → reviewed production merge", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "background", "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "mixed" }, "claim_ids": [ "ramp-inspect--summary", "ramp-inspect--headline-metric", "ramp-inspect--architecture-sandbox", "ramp-inspect--architecture-harness", "ramp-inspect--architecture-model", "ramp-inspect--architecture-interfaces", "ramp-inspect--architecture-tool-access", "ramp-inspect--architecture-knowledge", "ramp-inspect--architecture-credentials", "ramp-inspect--primitives-0", "ramp-inspect--primitives-1", "ramp-inspect--primitives-2", "ramp-inspect--key-metrics-0", "ramp-inspect--key-metrics-1", "ramp-inspect--key-metrics-2", "ramp-inspect--key-metrics-3", "ramp-inspect--key-metrics-4", "ramp-inspect--key-metrics-5", "ramp-inspect--key-metrics-6", "ramp-inspect--key-metrics-7", "ramp-inspect--key-metrics-8", "ramp-inspect--key-metrics-9", "ramp-inspect--key-metrics-10", "ramp-inspect--lessons-learned-0", "ramp-inspect--lessons-learned-1", "ramp-inspect--lessons-learned-2", "ramp-inspect--lessons-learned-3", "ramp-inspect--lessons-learned-4", "ramp-inspect--operating-models-0" ], "source_ids": [ "ramp-inspect-source-1", "ramp-inspect-source-2", "ramp-inspect-source-3", "ramp-inspect-source-4", "ramp-inspect-source-5" ], "interfaces": [ "slack", "web", "chrome-extension", "github" ] }, { "id": "replit-manager-agent", "company": "Replit", "agent_name": "Manager agent (agent-of-agents)", "approach_type": "orchestration-system", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "replit-manager-agent-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review", "support", "research", "data" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "objective → verifiable multi-agent work product", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "replit-manager-agent--summary", "replit-manager-agent--headline-metric", "replit-manager-agent--architecture-sandbox", "replit-manager-agent--architecture-harness", "replit-manager-agent--architecture-model", "replit-manager-agent--architecture-interfaces", "replit-manager-agent--architecture-tool-access", "replit-manager-agent--architecture-context-mgmt", "replit-manager-agent--primitives-0", "replit-manager-agent--key-metrics-0", "replit-manager-agent--key-metrics-1", "replit-manager-agent--lessons-learned-0", "replit-manager-agent--lessons-learned-1", "replit-manager-agent--operating-models-0" ], "source_ids": [ "replit-manager-agent-source-1" ], "interfaces": [ "slack" ] }, { "id": "retool-retoolgpt", "company": "Retool", "agent_name": "RetoolGPT", "approach_type": "task-agent", "deployment_stage": "deployed", "year": 2025, "first_public_evidence": { "date": "2025-08", "source_id": "retool-retoolgpt-source-1" }, "last_reviewed_at": "2026-08-31", "status": "internal", "domains": [ "support", "coding" ], "autonomy": "assistive", "operating_models": [ { "scope": "internal question → sourced answer", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "retool-retoolgpt--summary", "retool-retoolgpt--architecture-model", "retool-retoolgpt--architecture-knowledge", "retool-retoolgpt--operating-models-0" ], "source_ids": [ "retool-retoolgpt-source-1", "retool-retoolgpt-source-2" ], "interfaces": [] }, { "id": "salesforce-slackbot", "company": "Salesforce", "agent_name": "Slackbot", "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025-12-01", "source_id": "salesforce-slackbot-source-2" }, "last_reviewed_at": "2026-08-31", "status": "commercialized", "domains": [ "support", "customer-success", "ops" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "employee request → drafted work", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "interactive" ], "state": "unknown", "identity": "user", "evidence_strength": "detailed-primary" }, "claim_ids": [ "salesforce-slackbot--summary", "salesforce-slackbot--architecture-harness", "salesforce-slackbot--architecture-model", "salesforce-slackbot--architecture-interfaces", "salesforce-slackbot--architecture-tool-access", "salesforce-slackbot--architecture-knowledge", "salesforce-slackbot--architecture-credentials", "salesforce-slackbot--primitives-0", "salesforce-slackbot--lessons-learned-0", "salesforce-slackbot--lessons-learned-1", "salesforce-slackbot--operating-models-0" ], "source_ids": [ "salesforce-slackbot-source-1", "salesforce-slackbot-source-2", "salesforce-slackbot-source-3", "salesforce-slackbot-source-4" ], "interfaces": [ "slack" ] }, { "id": "sentry-junior", "company": "Sentry", "agent_name": "Junior", "approach_type": "task-agent", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "sentry-junior-source-2" }, "last_reviewed_at": "2026-09-09", "status": "open-sourced", "domains": [ "coding", "code-review", "support", "on-call" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "assigned task → human-steered and reviewed output", "attention_boundary": "continuous-steering", "level": 2 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "sentry-junior--summary", "sentry-junior--headline-metric", "sentry-junior--architecture-sandbox", "sentry-junior--architecture-harness", "sentry-junior--architecture-model", "sentry-junior--architecture-interfaces", "sentry-junior--architecture-tool-access", "sentry-junior--architecture-knowledge", "sentry-junior--architecture-credentials", "sentry-junior--architecture-context-mgmt", "sentry-junior--primitives-0", "sentry-junior--primitives-1", "sentry-junior--primitives-2", "sentry-junior--key-metrics-0", "sentry-junior--key-metrics-1", "sentry-junior--lessons-learned-0", "sentry-junior--lessons-learned-1", "sentry-junior--lessons-learned-2", "sentry-junior--lessons-learned-3", "sentry-junior--lessons-learned-4", "sentry-junior--operating-models-0" ], "source_ids": [ "sentry-junior-source-1", "sentry-junior-source-2", "sentry-junior-source-3", "sentry-junior-source-4" ], "interfaces": [ "slack", "web", "github" ] }, { "id": "shopify-internal-agents", "company": "Shopify", "agent_name": "Aquifer / River", "approach_type": "platform", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "shopify-internal-agents-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review", "research", "security" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "River coding request → reviewed pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "shopify-internal-agents--summary", "shopify-internal-agents--headline-metric", "shopify-internal-agents--architecture-sandbox", "shopify-internal-agents--architecture-harness", "shopify-internal-agents--architecture-model", "shopify-internal-agents--architecture-interfaces", "shopify-internal-agents--architecture-tool-access", "shopify-internal-agents--architecture-knowledge", "shopify-internal-agents--architecture-credentials", "shopify-internal-agents--architecture-context-mgmt", "shopify-internal-agents--primitives-0", "shopify-internal-agents--primitives-1", "shopify-internal-agents--primitives-2", "shopify-internal-agents--key-metrics-0", "shopify-internal-agents--key-metrics-1", "shopify-internal-agents--key-metrics-2", "shopify-internal-agents--lessons-learned-0", "shopify-internal-agents--lessons-learned-1", "shopify-internal-agents--lessons-learned-2", "shopify-internal-agents--lessons-learned-3", "shopify-internal-agents--operating-models-0" ], "source_ids": [ "shopify-internal-agents-source-1" ], "interfaces": [ "slack", "github" ] }, { "id": "sierra-pinecone", "company": "Sierra", "agent_name": "Pinecone", "approach_type": "task-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "sierra-pinecone-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review", "support", "research", "data" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "employee request → reviewed agent output", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "durable-session", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "sierra-pinecone--summary", "sierra-pinecone--headline-metric", "sierra-pinecone--architecture-sandbox", "sierra-pinecone--architecture-harness", "sierra-pinecone--architecture-model", "sierra-pinecone--architecture-interfaces", "sierra-pinecone--architecture-tool-access", "sierra-pinecone--architecture-knowledge", "sierra-pinecone--architecture-credentials", "sierra-pinecone--architecture-context-mgmt", "sierra-pinecone--primitives-0", "sierra-pinecone--primitives-1", "sierra-pinecone--primitives-2", "sierra-pinecone--key-metrics-0", "sierra-pinecone--key-metrics-1", "sierra-pinecone--lessons-learned-0", "sierra-pinecone--lessons-learned-1", "sierra-pinecone--lessons-learned-2", "sierra-pinecone--lessons-learned-3", "sierra-pinecone--operating-models-0" ], "source_ids": [ "sierra-pinecone-source-1", "sierra-pinecone-source-2", "sierra-pinecone-source-3" ], "interfaces": [ "slack", "web", "linear", "mobile" ] }, { "id": "slack-context-system", "company": "Slack", "agent_name": "Multi-agent context system", "approach_type": "supporting-pattern", "deployment_stage": "research", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "slack-context-system-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "research" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "long-running investigation → synthesized report", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "unknown" ], "state": "durable-session", "identity": "unknown", "evidence_strength": "mixed" }, "claim_ids": [ "slack-context-system--summary", "slack-context-system--headline-metric", "slack-context-system--architecture-harness", "slack-context-system--architecture-model", "slack-context-system--architecture-tool-access", "slack-context-system--architecture-context-mgmt", "slack-context-system--primitives-0", "slack-context-system--primitives-1", "slack-context-system--primitives-2", "slack-context-system--key-metrics-0", "slack-context-system--lessons-learned-0", "slack-context-system--lessons-learned-1", "slack-context-system--lessons-learned-2", "slack-context-system--operating-models-0" ], "source_ids": [ "slack-context-system-source-1", "slack-context-system-source-2" ], "interfaces": [] }, { "id": "spotify-honk-xirp", "company": "Spotify", "agent_name": "Honk / Xirp", "approach_type": "agent-system", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025", "source_id": "spotify-honk-xirp-source-1" }, "last_reviewed_at": "2026-09-09", "status": "mixed", "domains": [ "coding", "migrations", "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "Honk coding task → verified pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "durable-session", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "spotify-honk-xirp--summary", "spotify-honk-xirp--headline-metric", "spotify-honk-xirp--architecture-sandbox", "spotify-honk-xirp--architecture-harness", "spotify-honk-xirp--architecture-model", "spotify-honk-xirp--architecture-interfaces", "spotify-honk-xirp--architecture-tool-access", "spotify-honk-xirp--architecture-knowledge", "spotify-honk-xirp--architecture-credentials", "spotify-honk-xirp--architecture-context-mgmt", "spotify-honk-xirp--primitives-0", "spotify-honk-xirp--primitives-1", "spotify-honk-xirp--primitives-2", "spotify-honk-xirp--key-metrics-0", "spotify-honk-xirp--key-metrics-1", "spotify-honk-xirp--key-metrics-2", "spotify-honk-xirp--lessons-learned-0", "spotify-honk-xirp--lessons-learned-1", "spotify-honk-xirp--lessons-learned-2", "spotify-honk-xirp--lessons-learned-3", "spotify-honk-xirp--operating-models-0" ], "source_ids": [ "spotify-honk-xirp-source-1", "spotify-honk-xirp-source-2", "spotify-honk-xirp-source-3" ], "interfaces": [ "slack", "github", "cli" ] }, { "id": "stripe-minions", "company": "Stripe", "agent_name": "Minions", "approach_type": "background-agent", "deployment_stage": "scaled", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "stripe-minions-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "work context → merge-ready pull request", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "background", "event-driven", "interactive" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "stripe-minions--summary", "stripe-minions--headline-metric", "stripe-minions--architecture-sandbox", "stripe-minions--architecture-harness", "stripe-minions--architecture-model", "stripe-minions--architecture-interfaces", "stripe-minions--architecture-tool-access", "stripe-minions--architecture-knowledge", "stripe-minions--architecture-credentials", "stripe-minions--primitives-0", "stripe-minions--key-metrics-0", "stripe-minions--lessons-learned-0", "stripe-minions--lessons-learned-1", "stripe-minions--operating-models-0" ], "source_ids": [ "stripe-minions-source-1", "stripe-minions-source-2", "stripe-minions-source-3", "stripe-minions-source-4", "stripe-minions-source-5", "stripe-minions-source-6", "stripe-minions-source-7" ], "interfaces": [ "slack", "github", "cli", "web", "internal-ui" ] }, { "id": "uber-coding-agent", "company": "Uber", "agent_name": "Internal coding agent (unnamed)", "approach_type": "task-agent", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "uber-coding-agent-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "coding request → complete code change", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "unknown" ], "state": "unknown", "identity": "unknown", "evidence_strength": "secondary-only" }, "claim_ids": [ "uber-coding-agent--summary", "uber-coding-agent--headline-metric", "uber-coding-agent--architecture-sandbox", "uber-coding-agent--architecture-harness", "uber-coding-agent--key-metrics-0", "uber-coding-agent--key-metrics-1", "uber-coding-agent--operating-models-0" ], "source_ids": [ "uber-coding-agent-source-1" ], "interfaces": [] }, { "id": "uber-ureview", "company": "Uber", "agent_name": "uReview", "relationships": [ { "type": "related-to", "approach_id": "uber-coding-agent" } ], "approach_type": "background-agent", "deployment_stage": "scaled", "year": 2025, "first_public_evidence": { "date": "2025-08-12", "source_id": "uber-ureview-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "code-review" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "pull request → filtered AI review findings", "attention_boundary": "work-product-review", "level": 3 } ], "rubric": { "invocation": [ "background", "event-driven" ], "state": "unknown", "identity": "service", "evidence_strength": "detailed-primary" }, "claim_ids": [ "uber-ureview--summary", "uber-ureview--headline-metric", "uber-ureview--architecture-harness", "uber-ureview--architecture-model", "uber-ureview--architecture-interfaces", "uber-ureview--architecture-tool-access", "uber-ureview--architecture-knowledge", "uber-ureview--architecture-context-mgmt", "uber-ureview--primitives-0", "uber-ureview--primitives-1", "uber-ureview--primitives-2", "uber-ureview--key-metrics-0", "uber-ureview--key-metrics-1", "uber-ureview--key-metrics-2", "uber-ureview--key-metrics-3", "uber-ureview--key-metrics-4", "uber-ureview--lessons-learned-0", "uber-ureview--lessons-learned-1", "uber-ureview--lessons-learned-2", "uber-ureview--operating-models-0" ], "source_ids": [ "uber-ureview-source-1" ], "interfaces": [ "internal-ui", "ci" ] }, { "id": "workos-project-horizon", "company": "WorkOS", "agent_name": "Project Horizon", "approach_type": "platform", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "workos-project-horizon-source-1" }, "last_reviewed_at": "2026-08-31", "status": "internal", "domains": [ "coding", "code-review", "security" ], "autonomy": "drafts-reviewed", "operating_models": [ { "scope": "requirements and acceptance criteria → tested implementation", "attention_boundary": "outcome-review", "level": 4 } ], "rubric": { "invocation": [ "event-driven", "interactive" ], "state": "durable-session", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "workos-project-horizon--summary", "workos-project-horizon--architecture-sandbox", "workos-project-horizon--architecture-harness", "workos-project-horizon--architecture-model", "workos-project-horizon--architecture-interfaces", "workos-project-horizon--architecture-tool-access", "workos-project-horizon--architecture-knowledge", "workos-project-horizon--architecture-credentials", "workos-project-horizon--architecture-context-mgmt", "workos-project-horizon--primitives-0", "workos-project-horizon--primitives-1", "workos-project-horizon--primitives-2", "workos-project-horizon--primitives-3", "workos-project-horizon--lessons-learned-0", "workos-project-horizon--lessons-learned-1", "workos-project-horizon--lessons-learned-2", "workos-project-horizon--lessons-learned-3", "workos-project-horizon--lessons-learned-4", "workos-project-horizon--operating-models-0" ], "source_ids": [ "workos-project-horizon-source-1", "workos-project-horizon-source-2", "workos-project-horizon-source-3", "workos-project-horizon-source-4" ], "interfaces": [ "linear", "github", "slack", "web" ] }, { "id": "ycombinator-agent-infra", "company": "Y Combinator", "agent_name": "Internal agent infrastructure", "approach_type": "platform", "deployment_stage": "deployed", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "ycombinator-agent-infra-source-1" }, "last_reviewed_at": "2026-09-09", "status": "internal", "domains": [ "coding", "ops" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "internal request → agent-assisted organizational work", "attention_boundary": "unknown", "level": null } ], "rubric": { "invocation": [ "unknown" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "ycombinator-agent-infra--summary", "ycombinator-agent-infra--architecture-sandbox", "ycombinator-agent-infra--architecture-harness", "ycombinator-agent-infra--lessons-learned-0", "ycombinator-agent-infra--lessons-learned-1", "ycombinator-agent-infra--operating-models-0" ], "source_ids": [ "ycombinator-agent-infra-source-1" ], "interfaces": [] }, { "id": "zup-codegen", "company": "Zup", "agent_name": "CodeGen", "approach_type": "task-agent", "deployment_stage": "research", "year": 2026, "first_public_evidence": { "date": "2026", "source_id": "zup-codegen-source-1" }, "last_reviewed_at": "2026-08-31", "status": "internal", "domains": [ "coding" ], "autonomy": "human-in-loop", "operating_models": [ { "scope": "constrained coding task → human-supervised edit", "attention_boundary": "continuous-steering", "level": 2 } ], "rubric": { "invocation": [ "unknown" ], "state": "unknown", "identity": "unknown", "evidence_strength": "detailed-primary" }, "claim_ids": [ "zup-codegen--summary", "zup-codegen--architecture-harness", "zup-codegen--architecture-model", "zup-codegen--architecture-tool-access", "zup-codegen--architecture-credentials", "zup-codegen--primitives-0", "zup-codegen--primitives-1", "zup-codegen--lessons-learned-0", "zup-codegen--lessons-learned-1", "zup-codegen--operating-models-0" ], "source_ids": [ "zup-codegen-source-1" ], "interfaces": [] } ], "claims": [ { "id": "airbnb-airchat--summary", "approach_id": "airbnb-airchat", "field": "summary", "text": "Airbnb's internal agentic-coding harness, built by its Dev AI team. Airchat is a wrapper over Claude Code with a unified gateway for cost and metrics, an internal plugin marketplace, AirDev Workspaces for parallel sessions, and more than a dozen internal MCP servers that connect agents to internal systems. The team abandoned an earlier from-scratch orchestrator and shipped a thin shim over Airchat instead.", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "The build is described by Airbnb engineers in talks and podcasts, not in a first-party engineering blog.", "valid_at": null, "evidence": [ { "source_id": "airbnb-airchat-source-1", "relation": "contextualizes" }, { "source_id": "airbnb-airchat-source-2", "relation": "supports" } ] }, { "id": "airbnb-airchat--headline-metric", "approach_id": "airbnb-airchat", "field": "headline_metric", "text": "About 64% of pull requests materialized through agentic coding", "kind": "metric", "provenance": "reported", "confidence": "low", "confidence_reason": "The 64% figure comes from a third-party newsletter that quotes the engineers, not from a first-party Airbnb source.", "valid_at": "2025-10", "evidence": [ { "source_id": "airbnb-airchat-source-3", "relation": "supports", "locator": "Preserved content.md, lines 16, 24, 194" } ], "reported_by": "Airbnb", "metric_scope": "Airbnb PRs materialized through agentic coding by the October 2025 talk", "denominator": "Airbnb pull requests; exact count not supplied" }, { "id": "airbnb-airchat--architecture-harness", "approach_id": "airbnb-airchat", "field": "architecture.harness", "text": "Wrapper over Claude Code with a unified gateway, an internal plugin marketplace, and AirDev parallel workspaces", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "airbnb-airchat-source-2", "relation": "supports" } ] }, { "id": "airbnb-airchat--architecture-model", "approach_id": "airbnb-airchat", "field": "architecture.model", "text": "Claude Code (a vendor agent), wrapped by Airbnb", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "airbnb-airchat-source-2", "relation": "supports" } ] }, { "id": "airbnb-airchat--architecture-tool-access", "approach_id": "airbnb-airchat", "field": "architecture.tool_access", "text": "More than a dozen internal MCP servers connect agents to internal systems", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "airbnb-airchat-source-2", "relation": "supports" } ] }, { "id": "airbnb-airchat--key-metrics-0", "approach_id": "airbnb-airchat", "field": "key_metrics.0", "text": "About 64% of pull requests materialized through agentic coding", "kind": "metric", "provenance": "reported", "confidence": "low", "confidence_reason": "The figure comes from a third-party newsletter, not a first-party Airbnb source.", "valid_at": "2025-10", "evidence": [ { "source_id": "airbnb-airchat-source-3", "relation": "supports", "locator": "Preserved content.md, lines 16, 24, 194" } ], "reported_by": "Airbnb", "metric_scope": "Airbnb PRs materialized through agentic coding by the October 2025 talk", "denominator": "Airbnb pull requests; exact count not supplied" }, { "id": "airbnb-airchat--operating-models-0", "approach_id": "airbnb-airchat", "field": "operating_models.0", "text": "Level 3 for coding task → reviewed pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "Airbnb engineers describe agents producing pull requests that engineers review, which locates human attention at work-product review.", "valid_at": "2025", "evidence": [ { "source_id": "airbnb-airchat-source-2", "relation": "supports" } ] }, { "id": "atlassian-rovo-dev--summary", "approach_id": "atlassian-rovo-dev", "field": "summary", "text": "Atlassian's internal coding agent, built on the HULA (Human-in-the-loop software development agents) framework. Rovo Dev works inside Jira and runs a four-step cycle (set context, generate a plan, generate code, and raise a pull request). Atlassian dogfooded it across all Jira sites for more than a year across 1,900+ repositories. It reached general availability in October 2025.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "atlassian-rovo-dev-source-1", "relation": "supports" }, { "source_id": "atlassian-rovo-dev-source-3", "relation": "supports" } ] }, { "id": "atlassian-rovo-dev--headline-metric", "approach_id": "atlassian-rovo-dev", "field": "headline_metric", "text": "Dogfooded across 1,900+ repositories with a 50,000+ comment internal dataset", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Atlassian reported the dogfooding scale in its own engineering blog without independent verification.", "valid_at": "2026", "evidence": [ { "source_id": "atlassian-rovo-dev-source-2", "relation": "supports", "locator": "Preserved content.md, lines 66, 87" } ], "reported_by": "Atlassian", "metric_scope": "Internal code-review dogfooding repositories and Rovo Dev-generated classifier training comments" }, { "id": "atlassian-rovo-dev--architecture-harness", "approach_id": "atlassian-rovo-dev", "field": "architecture.harness", "text": "Built on the HULA framework, which runs set context, generate plan, generate code, and raise PR", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "atlassian-rovo-dev-source-1", "relation": "supports" } ] }, { "id": "atlassian-rovo-dev--architecture-interfaces", "approach_id": "atlassian-rovo-dev", "field": "architecture.interfaces", "text": "jira", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "atlassian-rovo-dev-source-1", "relation": "supports" } ] }, { "id": "atlassian-rovo-dev--key-metrics-0", "approach_id": "atlassian-rovo-dev", "field": "key_metrics.0", "text": "Dogfooded across 1,900+ repositories over more than a year", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Atlassian reported the dogfooding scale in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "atlassian-rovo-dev-source-2", "relation": "supports", "locator": "Preserved content.md, lines 87" } ], "reported_by": "Atlassian", "metric_scope": "Internal code-review evaluation across 1,900+ repositories spanning over a year", "measurement_method": "Year-long online evaluation across internal repositories" }, { "id": "atlassian-rovo-dev--key-metrics-1", "approach_id": "atlassian-rovo-dev", "field": "key_metrics.1", "text": "Trained on a proprietary internal dogfooding dataset of 50,000+ Rovo Dev comments", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Atlassian reported the dataset size in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "atlassian-rovo-dev-source-2", "relation": "supports", "locator": "Preserved content.md, lines 66" } ], "reported_by": "Atlassian", "metric_scope": "ModernBERT classifier training dataset of internally sourced Rovo Dev-generated comments", "measurement_method": "Internal dogfooding comments labeled by whether they led to a code resolution" }, { "id": "atlassian-rovo-dev--operating-models-0", "approach_id": "atlassian-rovo-dev", "field": "operating_models.0", "text": "Level 3 for Jira issue → reviewed pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The HULA framework and engineering blog describe a human-in-the-loop cycle that ends in a reviewed pull request.", "valid_at": "2024-11", "evidence": [ { "source_id": "atlassian-rovo-dev-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--summary", "approach_id": "block-builderbot", "field": "summary", "text": "A multi-agent orchestration layer built on goose + MCP that coordinates agents across Block's entire codebase, invoked in Slack to take a ticket end-to-end to a reviewed PR.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" }, { "source_id": "block-builderbot-source-2", "relation": "supports" }, { "source_id": "block-builderbot-source-3", "relation": "contextualizes" }, { "source_id": "block-builderbot-source-4", "relation": "contextualizes" } ] }, { "id": "block-builderbot--headline-metric", "approach_id": "block-builderbot", "field": "headline_metric", "text": "~1,500 PRs merged per week (~15% of all production code changes at Block)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24" } ], "reported_by": "Block", "metric_scope": "Builderbot merged pull requests per week at Block", "denominator": "All production code changes across Block for the approximately 15% share" }, { "id": "block-builderbot--architecture-sandbox", "approach_id": "block-builderbot", "field": "architecture.sandbox", "text": "unknown", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The preserved sources do not document an execution sandbox; unknown does not mean absent.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--architecture-harness", "approach_id": "block-builderbot", "field": "architecture.harness", "text": "Multi-agent orchestration built on goose (open-source agent framework) + MCP; multi-player, real-time, operating inside Slack threads", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--architecture-model", "approach_id": "block-builderbot", "field": "architecture.model", "text": "goose framework; model not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--architecture-interfaces", "approach_id": "block-builderbot", "field": "architecture.interfaces", "text": "slack, linear, jira, github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--architecture-tool-access", "approach_id": "block-builderbot", "field": "architecture.tool_access", "text": "MCP connects agents to internal tools and data; picks up Linear/Jira tickets, creates the branch, writes code, opens the PR, watches CI", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--architecture-knowledge", "approach_id": "block-builderbot", "field": "architecture.knowledge", "text": "Company-wide code context across hundreds of millions of lines and hundreds of services; Block also frames Builderbot as an 'agentic protector' around its software world model", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" }, { "source_id": "block-builderbot-source-2", "relation": "supports" } ] }, { "id": "block-builderbot--primitives-0", "approach_id": "block-builderbot", "field": "primitives.0", "text": "Coordinates multiple agents over one codebase instead of running a single loop", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--primitives-1", "approach_id": "block-builderbot", "field": "primitives.1", "text": "Linear/Jira ticket -> branch -> code -> PR -> CI watch, end to end", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--key-metrics-0", "approach_id": "block-builderbot", "field": "key_metrics.0", "text": "200,000+ operations per day", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24" } ], "reported_by": "Block", "metric_scope": "Builderbot operations per day; the article does not define an operation" }, { "id": "block-builderbot--key-metrics-1", "approach_id": "block-builderbot", "field": "key_metrics.1", "text": "~1,500 pull requests merged per week (~15% of all production code changes at Block)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24" } ], "reported_by": "Block", "metric_scope": "Builderbot merged pull requests per week at Block", "denominator": "All production code changes across Block for the approximately 15% share" }, { "id": "block-builderbot--key-metrics-2", "approach_id": "block-builderbot", "field": "key_metrics.2", "text": "\"What used to take months now takes days\"", "kind": "opinion", "provenance": "reported", "confidence": "low", "confidence_reason": "Qualitative company statement, not a measured before-and-after study.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24–26" } ], "metric_scope": "Reported turnaround for Square seller features and repetitive engineering work" }, { "id": "block-builderbot--lessons-learned-0", "approach_id": "block-builderbot", "field": "lessons_learned.0", "text": "Concentrate investment on orchestration, context, and the environment; let engineers focus on the problems worth solving", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--lessons-learned-1", "approach_id": "block-builderbot", "field": "lessons_learned.1", "text": "Meet people in Slack: tag @builderbot with a short description and it works in the thread", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--lessons-learned-2", "approach_id": "block-builderbot", "field": "lessons_learned.2", "text": "Multi-player real-time collaboration lets humans steer research, planning, and implementation rather than only reviewing after the fact", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "block-builderbot--operating-models-0", "approach_id": "block-builderbot", "field": "operating_models.0", "text": "Level 3 for ticket → reviewed pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source explicitly describes end-to-end implementation ending in a reviewed pull request.", "valid_at": "2026-06-21", "evidence": [ { "source_id": "block-builderbot-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--summary", "approach_id": "brex-agent-platform", "field": "summary", "text": "Retool-based internal platform where employees build, test, and deploy agents for KYC, disputes, QA, collections, and operations.", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--headline-metric", "approach_id": "brex-agent-platform", "field": "headline_metric", "text": "Dispute processing time fell from three hours to three seconds", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2025", "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports", "locator": "Preserved content.md, lines 179–189" } ], "reported_by": "Brex", "metric_scope": "Dispute-submission preparation using the internal agent platform; not end-to-end chargeback resolution" }, { "id": "brex-agent-platform--architecture-sandbox", "approach_id": "brex-agent-platform", "field": "architecture.sandbox", "text": "Retool-hosted runtime (no bespoke execution env described)", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--architecture-harness", "approach_id": "brex-agent-platform", "field": "architecture.harness", "text": "Retool-based builder with prompt management and multi-model testing/evaluation; built by a ~25-person systems-engineering team", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--architecture-model", "approach_id": "brex-agent-platform", "field": "architecture.model", "text": "Multi-model", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--architecture-interfaces", "approach_id": "brex-agent-platform", "field": "architecture.interfaces", "text": "slack, internal-ui", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--architecture-tool-access", "approach_id": "brex-agent-platform", "field": "architecture.tool_access", "text": "An MCP server exposes external product features to the internal platform; new product tools become internally available immediately; invoked via Slack /c1", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--architecture-knowledge", "approach_id": "brex-agent-platform", "field": "architecture.knowledge", "text": "Standard operating procedures uploaded as a knowledge base, such as 100-page dispute guides; customer account data", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--architecture-credentials", "approach_id": "brex-agent-platform", "field": "architecture.credentials", "text": "SSO via internal Retool proxies (no per-user accounts); ConductorOne access management; Okta auth; data classified by risk; ≤30-day retention, no training on inputs", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--primitives-0", "approach_id": "brex-agent-platform", "field": "primitives.0", "text": "Non-technical ops staff design prompts, test across models, deploy with QA oversight", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--primitives-1", "approach_id": "brex-agent-platform", "field": "primitives.1", "text": "Product features exposed to internal agents through one MCP server", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--key-metrics-0", "approach_id": "brex-agent-platform", "field": "key_metrics.0", "text": "50%+ of customer-support cases resolved by chatbot as first touch", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2025", "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports", "locator": "Preserved content.md, lines 76" } ], "reported_by": "Brex", "metric_scope": "Customer-support cases resolved by the chatbot at first touch", "denominator": "Customer-support cases; exact sample size not provided" }, { "id": "brex-agent-platform--key-metrics-1", "approach_id": "brex-agent-platform", "field": "key_metrics.1", "text": "Dispute processing: 3 hours → 3 seconds", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2025", "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports", "locator": "Preserved content.md, lines 179–189" } ], "reported_by": "Brex", "metric_scope": "Dispute-submission preparation using the internal agent platform; not end-to-end chargeback resolution" }, { "id": "brex-agent-platform--key-metrics-2", "approach_id": "brex-agent-platform", "field": "key_metrics.2", "text": "QA covers every support interaction; one person using AI instead of five QA specialists", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2025", "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports", "locator": "Preserved content.md, lines 80" } ], "reported_by": "Brex", "metric_scope": "Quality assurance of customer-support interactions", "denominator": "Every support interaction", "measurement_method": "Agent applies the quality rubric to every response; one person oversees instead of five QA specialists" }, { "id": "brex-agent-platform--key-metrics-3", "approach_id": "brex-agent-platform", "field": "key_metrics.3", "text": "KYC adverse-media accuracy 85% → 88%", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2025", "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports", "locator": "Preserved content.md, lines 163–167" } ], "reported_by": "Brex", "metric_scope": "KYC adverse-media classification; human versus agent accuracy" }, { "id": "brex-agent-platform--lessons-learned-0", "approach_id": "brex-agent-platform", "field": "lessons_learned.0", "text": "Target 40% automation, not 100%; the last mile drives investments that often yield zero value", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--lessons-learned-1", "approach_id": "brex-agent-platform", "field": "lessons_learned.1", "text": "Brex reports that it manages the internal platform with practices used for an external product", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--lessons-learned-2", "approach_id": "brex-agent-platform", "field": "lessons_learned.2", "text": "Don't skip the human in the middle; end-to-end automation fails on accuracy", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--lessons-learned-3", "approach_id": "brex-agent-platform", "field": "lessons_learned.3", "text": "Map workflows to the discrete steps a human would take, then translate each to LLM instructions", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--lessons-learned-4", "approach_id": "brex-agent-platform", "field": "lessons_learned.4", "text": "Architecture beats vendor; legal/data approvals should follow data-handling characteristics, not the tool name", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "brex-agent-platform--operating-models-0", "approach_id": "brex-agent-platform", "field": "operating_models.0", "text": "Level 2 for internal operations request → completed operation; human attention boundary: continuous-steering.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The secondary report describes a human remaining in the middle of the workflow, but does not fully specify each review surface.", "valid_at": "2025-09-25", "evidence": [ { "source_id": "brex-agent-platform-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--summary", "approach_id": "browserbase-bb", "field": "summary", "text": "One generalized agent in Slack that writes PRs, investigates sessions, queries the warehouse, logs feature requests, and runs browser agents across engineering, ops, sales, and support.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--headline-metric", "approach_id": "browserbase-bb", "field": "headline_metric", "text": "Feature-request pipeline at 100% coverage with zero human effort", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports", "locator": "Preserved content.md, lines 26" } ], "reported_by": "Browserbase", "metric_scope": "Automatic feature-request scanning of every closed support ticket and meeting transcript", "denominator": "Closed support tickets and meeting transcripts" }, { "id": "browserbase-bb--architecture-sandbox", "approach_id": "browserbase-bb", "field": "architecture.sandbox", "text": "Ephemeral Linux VM; pre-warmed snapshot rebuilt every 30 min; idles out after 30 min", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--architecture-harness", "approach_id": "browserbase-bb", "field": "architecture.harness", "text": "OpenCode core loop with 6 tools (read/write/edit/exec/safebash/skill)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--architecture-model", "approach_id": "browserbase-bb", "field": "architecture.model", "text": "Frontier models", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--architecture-interfaces", "approach_id": "browserbase-bb", "field": "architecture.interfaces", "text": "slack, web, webhook", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--architecture-tool-access", "approach_id": "browserbase-bb", "field": "architecture.tool_access", "text": "exec routes through a serverless integration proxy (Snowflake, HubSpot, Pylon, Grafana); the sandbox never sees real secrets", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--architecture-knowledge", "approach_id": "browserbase-bb", "field": "architecture.knowledge", "text": "Key repos cloned into /knowledge/; skills (markdown) loaded on demand", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--architecture-credentials", "approach_id": "browserbase-bb", "field": "architecture.credentials", "text": "Credential brokering; the sandbox boots with references + rotating session tokens only; the proxy holds real creds; egress injection for a few hosts", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--primitives-0", "approach_id": "browserbase-bb", "field": "primitives.0", "text": "Markdown playbooks lazy-loaded per task so the general agent stays small", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--primitives-1", "approach_id": "browserbase-bb", "field": "primitives.1", "text": "Scoped permissions per session limit the actions available to the agent", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--primitives-2", "approach_id": "browserbase-bb", "field": "primitives.2", "text": "The sandbox runs arbitrary code yet never touches a secret", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--key-metrics-0", "approach_id": "browserbase-bb", "field": "key_metrics.0", "text": "Feature-request pipeline at 100% coverage, zero human effort", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports", "locator": "Preserved content.md, lines 26" } ], "reported_by": "Browserbase", "metric_scope": "Automatic feature-request scanning of every closed support ticket and meeting transcript", "denominator": "Closed support tickets and meeting transcripts" }, { "id": "browserbase-bb--key-metrics-1", "approach_id": "browserbase-bb", "field": "key_metrics.1", "text": "99% of first-response times < 24 hrs", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "The source reports 99% below 24 hours with awkward wording; no sample, timestamps, or method is provided.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports", "locator": "Preserved content.md, lines 26" } ], "reported_by": "Browserbase", "metric_scope": "Reported support first-response time below 24 hours", "denominator": "Support first responses; exact sample and measurement window not provided" }, { "id": "browserbase-bb--key-metrics-2", "approach_id": "browserbase-bb", "field": "key_metrics.2", "text": "Session investigation: 30–60 min of log-diving → one Slack message", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A Slack message replaces the manual initiation workflow; the article does not report end-to-end automated investigation latency.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports", "locator": "Preserved content.md, lines 26" } ], "reported_by": "Browserbase", "metric_scope": "Session investigation initiation: manual log-diving versus a Slack request" }, { "id": "browserbase-bb--lessons-learned-0", "approach_id": "browserbase-bb", "field": "lessons_learned.0", "text": "One agent with good abstractions beats a fleet of narrow bots", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--lessons-learned-1", "approach_id": "browserbase-bb", "field": "lessons_learned.1", "text": "Separate capabilities from the core loop; domain logic lives in skills + service packages", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--lessons-learned-2", "approach_id": "browserbase-bb", "field": "lessons_learned.2", "text": "Don't trust the model, remove its ability to do wrong: scope tools and services per invocation source", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--lessons-learned-3", "approach_id": "browserbase-bb", "field": "lessons_learned.3", "text": "Meet people where they are; Slack is the highest-leverage surface because that's where work already happens", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "browserbase-bb--operating-models-0", "approach_id": "browserbase-bb", "field": "operating_models.0", "text": "Level 3 for coding request → reviewed pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source documents agent-authored pull requests while the enclosing workflow retains human review.", "valid_at": "2026", "evidence": [ { "source_id": "browserbase-bb-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--summary", "approach_id": "cloudflare-ai-stack", "field": "summary", "text": "An internal platform of MCP servers, an access layer, and AI tooling (incl. an AI code reviewer) that makes agents useful inside Cloudflare.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" }, { "source_id": "cloudflare-ai-stack-source-2", "relation": "contextualizes" } ] }, { "id": "cloudflare-ai-stack--headline-metric", "approach_id": "cloudflare-ai-stack", "field": "headline_metric", "text": "47.95 million AI requests in 30 days across the internal AI engineering system", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026", "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14–20" } ], "reported_by": "Cloudflare", "metric_scope": "Internal AI engineering requests in the 30 days preceding the report", "measurement_method": "Company-reported AI Gateway count for the preceding 30 days" }, { "id": "cloudflare-ai-stack--architecture-sandbox", "approach_id": "cloudflare-ai-stack", "field": "architecture.sandbox", "text": "Dynamic Workers for sandboxed code execution; Sandbox SDK to clone/build/test", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--architecture-harness", "approach_id": "cloudflare-ai-stack", "field": "architecture.harness", "text": "OpenCode + Windsurf clients; Agents SDK (McpAgent + Durable Objects) for stateful sessions", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--architecture-model", "approach_id": "cloudflare-ai-stack", "field": "architecture.model", "text": "Workers AI (open-weight, on-platform) + frontier models (Opus, GPT), routed by task", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--architecture-interfaces", "approach_id": "cloudflare-ai-stack", "field": "architecture.interfaces", "text": "cli, ci, web", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--architecture-tool-access", "approach_id": "cloudflare-ai-stack", "field": "architecture.tool_access", "text": "MCP Server Portal; one OAuth point aggregating 182+ tools from 13 servers; AI Gateway for routing, cost, BYOK, ZDR", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--architecture-knowledge", "approach_id": "cloudflare-ai-stack", "field": "architecture.knowledge", "text": "Backstage catalog (2,055 services) + AGENTS.md generated across ~3,900 repos", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--architecture-credentials", "approach_id": "cloudflare-ai-stack", "field": "architecture.credentials", "text": "Zero API keys on client machines; a Worker injects keys server-side; Cloudflare Access (Zero Trust) auth", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--architecture-context-mgmt", "approach_id": "cloudflare-ai-stack", "field": "architecture.context_mgmt", "text": "Code Mode collapses upstream tool schemas into search + execute, holding token overhead constant at scale", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--primitives-0", "approach_id": "cloudflare-ai-stack", "field": "primitives.0", "text": "One OAuth aggregation point for all MCP tools", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--primitives-1", "approach_id": "cloudflare-ai-stack", "field": "primitives.1", "text": "Collapse N tool schemas into 2 calls to hold token overhead constant at scale", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--primitives-2", "approach_id": "cloudflare-ai-stack", "field": "primitives.2", "text": "Structured, generated repo context (runtime, nav, conventions, boundaries, deps)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--primitives-3", "approach_id": "cloudflare-ai-stack", "field": "primitives.3", "text": "Multi-agent CI review: risk tiering, specialist agents, Codex-rule citations", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--key-metrics-0", "approach_id": "cloudflare-ai-stack", "field": "key_metrics.0", "text": "3,683 internal users (60% of company, 93% of R&D)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026", "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14–16" } ], "reported_by": "Cloudflare", "metric_scope": "Active internal AI coding-tool users in the preceding 30 days", "denominator": "Approximately 6,100 employees for company share; R&D organization for R&D share" }, { "id": "cloudflare-ai-stack--key-metrics-1", "approach_id": "cloudflare-ai-stack", "field": "key_metrics.1", "text": "47.95M AI requests and 241.37B tokens via AI Gateway in the preceding 30 days", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026", "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14–20" } ], "reported_by": "Cloudflare", "metric_scope": "Internal AI requests and AI Gateway tokens in the preceding 30 days", "measurement_method": "Reported request counts and AI Gateway token counts" }, { "id": "cloudflare-ai-stack--key-metrics-2", "approach_id": "cloudflare-ai-stack", "field": "key_metrics.2", "text": "10,952 merge requests in the week of March 23, 2026, nearly double the Q4 baseline; four-week average above 8,700", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026-03", "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports", "locator": "Preserved content.md, lines 25–29" } ], "reported_by": "Cloudflare", "metric_scope": "Company merge requests in the week of March 23, 2026, versus Q4 baseline; not agent-authored PRs", "measurement_method": "Weekly merge-request count; distinct from the four-week rolling average" }, { "id": "cloudflare-ai-stack--key-metrics-3", "approach_id": "cloudflare-ai-stack", "field": "key_metrics.3", "text": "295 teams using agentic AI tools", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026", "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14–18" } ], "reported_by": "Cloudflare", "metric_scope": "Teams using agentic AI tools and coding assistants in the reported 30-day snapshot" }, { "id": "cloudflare-ai-stack--lessons-learned-0", "approach_id": "cloudflare-ai-stack", "field": "lessons_learned.0", "text": "Centralize through a proxy early; direct-to-gateway looks simpler but blocks per-user attribution, model cataloging, and policy later", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--lessons-learned-1", "approach_id": "cloudflare-ai-stack", "field": "lessons_learned.1", "text": "Without structured data, agents are working blind; they read code but can't see the system around it (Backstage / AGENTS.md)", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--lessons-learned-2", "approach_id": "cloudflare-ai-stack", "field": "lessons_learned.2", "text": "Tool schemas eat context (34 GitLab tools ≈ 7.5% of a 200K window); collapse them at the portal", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--lessons-learned-3", "approach_id": "cloudflare-ai-stack", "field": "lessons_learned.3", "text": "Frontier + open-source hybrid: route a growing share of workloads to cheaper self-hosted models", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "cloudflare-ai-stack--operating-models-0", "approach_id": "cloudflare-ai-stack", "field": "operating_models.0", "text": "Level 3 for pull request → AI review findings; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The source documents automated review findings, but the record combines several platform workflows.", "valid_at": "2026-04-20", "evidence": [ { "source_id": "cloudflare-ai-stack-source-1", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--summary", "approach_id": "coinbase-forge-mux", "field": "summary", "text": "Forge turns a Slack/GitHub/Linear discussion into a Linear issue, fix, PR, and one-off build; Mux lets employees run many coding agents concurrently.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" }, { "source_id": "coinbase-forge-mux-source-2", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--headline-metric", "approach_id": "coinbase-forge-mux", "field": "headline_metric", "text": "Mux: 600+ users including engineers, PMs, and designers (335 active, 197 power users)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24–34" } ], "reported_by": "Coinbase", "metric_scope": "Registered Mux users including engineers, PMs, and designers; 335 active and 197 power users" }, { "id": "coinbase-forge-mux--architecture-sandbox", "approach_id": "coinbase-forge-mux", "field": "architecture.sandbox", "text": "Mux gives each concurrent agent its own git worktree, branch, and terminal", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--architecture-harness", "approach_id": "coinbase-forge-mux", "field": "architecture.harness", "text": "Portfolio approach; Claude Code, OpenCode, Cursor, and Copilot rather than one harness; Forge is a custom harness invokable from Slack, GitHub, and Linear", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" }, { "source_id": "coinbase-forge-mux-source-2", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--architecture-model", "approach_id": "coinbase-forge-mux", "field": "architecture.model", "text": "Portfolio across multiple providers", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--architecture-interfaces", "approach_id": "coinbase-forge-mux", "field": "architecture.interfaces", "text": "slack, github, linear", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--architecture-tool-access", "approach_id": "coinbase-forge-mux", "field": "architecture.tool_access", "text": "Linear treated as the structured product context / source of truth", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-2", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--architecture-knowledge", "approach_id": "coinbase-forge-mux", "field": "architecture.knowledge", "text": "Linear as the durable structured-context layer for product work", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-2", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--primitives-0", "approach_id": "coinbase-forge-mux", "field": "primitives.0", "text": "Custom harness: Slack bug discussion → Linear issue → fix → PR → one-off build", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-2", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--primitives-1", "approach_id": "coinbase-forge-mux", "field": "primitives.1", "text": "Concurrency layer; one human coordinates many isolated coding agents, each in its own worktree/branch/terminal", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--key-metrics-0", "approach_id": "coinbase-forge-mux", "field": "key_metrics.0", "text": "Mux: 600+ users including engineers, PMs, and designers (335 active, 197 power users)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24–34" } ], "reported_by": "Coinbase", "metric_scope": "Registered Mux users including engineers, PMs, and designers; 335 active and 197 power users" }, { "id": "coinbase-forge-mux--lessons-learned-0", "approach_id": "coinbase-forge-mux", "field": "lessons_learned.0", "text": "Support a portfolio of harnesses (Claude Code, OpenCode, Cursor, Copilot) rather than standardizing on one", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--lessons-learned-1", "approach_id": "coinbase-forge-mux", "field": "lessons_learned.1", "text": "Keep the system of record (Linear) as the agent's structured context; conversation can be the input, Linear stays durable", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-2", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--lessons-learned-2", "approach_id": "coinbase-forge-mux", "field": "lessons_learned.2", "text": "Simple per-agent worktree/branch/terminal isolation is a pragmatic alternative to full sandboxing for concurrent coding", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" } ] }, { "id": "coinbase-forge-mux--operating-models-0", "approach_id": "coinbase-forge-mux", "field": "operating_models.0", "text": "Level 3 for Slack, GitHub, or Linear request → reviewed pull request and build; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source describes delegated implementation that returns a pull request and build for human review.", "valid_at": "2026", "evidence": [ { "source_id": "coinbase-forge-mux-source-1", "relation": "supports" } ] }, { "id": "databricks-costar--summary", "approach_id": "databricks-costar", "field": "summary", "text": "Databricks' internal engineering agents and the coSTAR framework that ships and tests them. Databricks uses internal agents as daily coding drivers on its own codebase, including code-review and on-call support work. coSTAR tests agents on a private benchmark built from Databricks' multi-million line codebase before they ship. Omnigent is a separate shipping open-source product and is excluded from this record.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "databricks-costar-source-1", "relation": "supports" }, { "source_id": "databricks-costar-source-2", "relation": "supports" } ] }, { "id": "databricks-costar--architecture-harness", "approach_id": "databricks-costar", "field": "architecture.harness", "text": "coSTAR framework for shipping and testing internal agents", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "databricks-costar-source-1", "relation": "supports" } ] }, { "id": "databricks-costar--architecture-knowledge", "approach_id": "databricks-costar", "field": "architecture.knowledge", "text": "Private benchmark built from the Databricks multi-million line codebase", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "databricks-costar-source-2", "relation": "supports" } ] }, { "id": "databricks-costar--key-metrics-0", "approach_id": "databricks-costar", "field": "key_metrics.0", "text": "Internal agents serve as daily coding drivers on the Databricks codebase", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "Databricks described internal use in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "databricks-costar-source-1", "relation": "supports" } ] }, { "id": "databricks-costar--key-metrics-1", "approach_id": "databricks-costar", "field": "key_metrics.1", "text": "Private benchmark built from a multi-million line codebase", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "Databricks described the benchmark in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "databricks-costar-source-2", "relation": "supports" } ] }, { "id": "databricks-costar--operating-models-0", "approach_id": "databricks-costar", "field": "operating_models.0", "text": "Unclassified for internal engineering workflows → agent-produced changes; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The record covers several internal engineering agents with different workflows, so no single human-attention boundary applies.", "valid_at": "2025", "evidence": [ { "source_id": "databricks-costar-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--summary", "approach_id": "domu-clementino", "field": "summary", "text": "A general-purpose 'AI colleague' spanning sales, finance, client ops, engineering, and recruitment, later split into a reusable toolkit plus Slack and desktop surfaces.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--headline-metric", "approach_id": "domu-clementino", "field": "headline_metric", "text": "~35 integrations organized into skills", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports", "locator": "Preserved content.md, lines 58–60" } ], "reported_by": "Domu", "metric_scope": "Clementino toolkit integration modules organized into skills" }, { "id": "domu-clementino--architecture-harness", "approach_id": "domu-clementino", "field": "architecture.harness", "text": "Claude/Anthropic SDK wrapper; a reusable tool/skill/memory layer separated from the Slack and desktop interfaces", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--architecture-model", "approach_id": "domu-clementino", "field": "architecture.model", "text": "Claude (Anthropic SDK)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--architecture-interfaces", "approach_id": "domu-clementino", "field": "architecture.interfaces", "text": "slack, desktop", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--architecture-tool-access", "approach_id": "domu-clementino", "field": "architecture.tool_access", "text": "~35 integrations organized into skills; specialist delegates per domain", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--architecture-knowledge", "approach_id": "domu-clementino", "field": "architecture.knowledge", "text": "Four memory layers: conversation context, persistent facts, knowledge RAG, and live system state", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--architecture-credentials", "approach_id": "domu-clementino", "field": "architecture.credentials", "text": "Customer-impacting actions gated behind team-visible Slack approvals", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--architecture-context-mgmt", "approach_id": "domu-clementino", "field": "architecture.context_mgmt", "text": "Four-layer memory separation; prompt size dropped substantially after splitting capabilities from the Slack layer", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--primitives-0", "approach_id": "domu-clementino", "field": "primitives.0", "text": "Transient conversation, persistent facts, knowledge RAG, and live system state; kept separate", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--primitives-1", "approach_id": "domu-clementino", "field": "primitives.1", "text": "Capabilities separated from the interface layer so they power multiple surfaces", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--key-metrics-0", "approach_id": "domu-clementino", "field": "key_metrics.0", "text": "~35 integrations organized into skills", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports", "locator": "Preserved content.md, lines 58–60" } ], "reported_by": "Domu", "metric_scope": "Clementino toolkit integration modules organized into skills" }, { "id": "domu-clementino--lessons-learned-0", "approach_id": "domu-clementino", "field": "lessons_learned.0", "text": "Separate reusable capabilities (tools/skills/memory) from the interface layer; prompt size drops while capability is preserved", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--lessons-learned-1", "approach_id": "domu-clementino", "field": "lessons_learned.1", "text": "Model memory explicitly: conversation context, persistent facts, RAG, and live state have different lifecycles", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--lessons-learned-2", "approach_id": "domu-clementino", "field": "lessons_learned.2", "text": "Gate customer-impacting actions behind human approvals rather than trusting the model", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "domu-clementino--operating-models-0", "approach_id": "domu-clementino", "field": "operating_models.0", "text": "Level 3 for employee request → approved customer-impacting action; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source documents explicit human approval for customer-impacting actions.", "valid_at": "2026", "evidence": [ { "source_id": "domu-clementino-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--summary", "approach_id": "doordash-code-review", "field": "summary", "text": "A specialized agent that automatically reviews 10,000+ PRs a week across 56 repositories, emphasizing grounded high-confidence findings over noisy comments.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--headline-metric", "approach_id": "doordash-code-review", "field": "headline_metric", "text": "10,000+ pull requests reviewed per week across 56 repositories", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports", "locator": "Preserved content.md, lines 23" } ], "reported_by": "DoorDash", "metric_scope": "Typical weekly PR reviews across 56 onboarded repositories" }, { "id": "doordash-code-review--architecture-harness", "approach_id": "doordash-code-review", "field": "architecture.harness", "text": "Three architecture versions; emphasis on attention and grounded, high-confidence findings rather than commenting everywhere", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--architecture-model", "approach_id": "doordash-code-review", "field": "architecture.model", "text": "Not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--architecture-interfaces", "approach_id": "doordash-code-review", "field": "architecture.interfaces", "text": "github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--architecture-tool-access", "approach_id": "doordash-code-review", "field": "architecture.tool_access", "text": "Reviews Go, iOS, Android, web, infrastructure, and data code", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--architecture-knowledge", "approach_id": "doordash-code-review", "field": "architecture.knowledge", "text": "Grounded findings tied to evidence", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--primitives-0", "approach_id": "doordash-code-review", "field": "primitives.0", "text": "High-confidence, evidence-backed comments rather than blanket commentary", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--key-metrics-0", "approach_id": "doordash-code-review", "field": "key_metrics.0", "text": "10,000+ PRs reviewed in a typical week across 56 repositories", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports", "locator": "Preserved content.md, lines 23" } ], "reported_by": "DoorDash", "metric_scope": "Typical weekly PR reviews across 56 onboarded repositories" }, { "id": "doordash-code-review--key-metrics-1", "approach_id": "doordash-code-review", "field": "key_metrics.1", "text": "60.2% action rate on settled high/critical findings (measured sample)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports", "locator": "Preserved content.md, lines 27" } ], "reported_by": "DoorDash", "metric_scope": "Settled high and critical findings that led to code changes before merge", "denominator": "2,256 settled high and critical findings", "measurement_method": "Whether the human changed the code before merge in response to the finding" }, { "id": "doordash-code-review--lessons-learned-0", "approach_id": "doordash-code-review", "field": "lessons_learned.0", "text": "Optimize for attention; minimize noisy comments; comment only with grounded, high-confidence findings", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--lessons-learned-1", "approach_id": "doordash-code-review", "field": "lessons_learned.1", "text": "Measure whether engineers actually act on findings (action rate), not comment volume", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-code-review--operating-models-0", "approach_id": "doordash-code-review", "field": "operating_models.0", "text": "Level 3 for pull request → AI review comments; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source describes automated findings that engineers evaluate within the pull-request workflow.", "valid_at": "2026", "evidence": [ { "source_id": "doordash-code-review-source-1", "relation": "supports" } ] }, { "id": "doordash-flux--summary", "approach_id": "doordash-flux", "field": "summary", "text": "DoorDash's internal agentic AI platform; a unified cognitive layer over company data and operations, with an AI Marketplace of specialized agents and the Flux cloud-agent runtime for engineering tasks.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-1", "relation": "supports" }, { "source_id": "doordash-flux-source-2", "relation": "supports" }, { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--headline-metric", "approach_id": "doordash-flux", "field": "headline_metric", "text": "130,000 engineering tasks automated in one month", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Dated August 11, 2026 report of a one-month count; this is the observation date, not the measurement window.", "valid_at": "2026-08-11", "evidence": [ { "source_id": "doordash-flux-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10–12" } ], "reported_by": "DoorDash", "metric_scope": "Engineering tasks automated in one reported month; calendar measurement month unspecified" }, { "id": "doordash-flux--architecture-sandbox", "approach_id": "doordash-flux", "field": "architecture.sandbox", "text": "Firecracker microVMs; <5s p95 end-to-end setup (boot, clone repos, install tools, configure harness)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--architecture-harness", "approach_id": "doordash-flux", "field": "architecture.harness", "text": "Maturity model: deterministic workflows -> ReAct agents -> hierarchical deep agents -> experimental swarms", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" } ] }, { "id": "doordash-flux--architecture-model", "approach_id": "doordash-flux", "field": "architecture.model", "text": "Model-agnostic platform primitives support third-party or in-house agent components", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--architecture-interfaces", "approach_id": "doordash-flux", "field": "architecture.interfaces", "text": "slack, github, scheduled, cli, skill, cursor", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--architecture-tool-access", "approach_id": "doordash-flux", "field": "architecture.tool_access", "text": "In-house MCP gateway ('Agent Gateway'); LangGraph orchestration; prospective A2A; tools declared per playbook with scoped, logged permissions", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" }, { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--architecture-knowledge", "approach_id": "doordash-flux", "field": "architecture.knowledge", "text": "AI Marketplace of specialized agents; DataExplorer for grounded analytics; DoorDash-specific context in playbooks", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" }, { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--architecture-credentials", "approach_id": "doordash-flux", "field": "architecture.credentials", "text": "Scoped per playbook; brokered through the gateway, never on the laptop; provenance on every action", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--architecture-context-mgmt", "approach_id": "doordash-flux", "field": "architecture.context_mgmt", "text": "Hybrid retrieval: BM25 + dense semantic + reciprocal-rank fusion -> RAG; schema-aware SQL with EXPLAIN validation", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" } ] }, { "id": "doordash-flux--primitives-0", "approach_id": "doordash-flux", "field": "primitives.0", "text": "Isolated Firecracker microVM with repos, tools, secrets, runtime deps", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--primitives-1", "approach_id": "doordash-flux", "field": "primitives.1", "text": "Governed, audited access to CI, observability, issue trackers, deploy, code search", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--primitives-2", "approach_id": "doordash-flux", "field": "primitives.2", "text": "YAML unit of agentic work: task, inputs, skills, tools, permissions, validation, outputs", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--primitives-3", "approach_id": "doordash-flux", "field": "primitives.3", "text": "Identifies schemas, generates grounded SQL, validates via EXPLAIN before execution", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" } ] }, { "id": "doordash-flux--primitives-4", "approach_id": "doordash-flux", "field": "primitives.4", "text": "Workflows -> agents -> deep-agent hierarchies -> swarms; governance hardens as control decentralizes", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" } ] }, { "id": "doordash-flux--key-metrics-0", "approach_id": "doordash-flux", "field": "key_metrics.0", "text": "130,000 engineering tasks automated in one month", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Dated August 11, 2026 report of a one-month count; this is the observation date, not the measurement window.", "valid_at": "2026-08-11", "evidence": [ { "source_id": "doordash-flux-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10–12" } ], "reported_by": "DoorDash", "metric_scope": "Engineering tasks automated in one reported month; calendar measurement month unspecified" }, { "id": "doordash-flux--key-metrics-1", "approach_id": "doordash-flux", "field": "key_metrics.1", "text": "25,000+ automated code reviews per week", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Dated August 11, 2026 report; weekly measurement boundaries are not supplied.", "valid_at": "2026-08-11", "evidence": [ { "source_id": "doordash-flux-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10–12" } ], "reported_by": "DoorDash", "metric_scope": "Weekly automated code reviews powered by Flux" }, { "id": "doordash-flux--key-metrics-2", "approach_id": "doordash-flux", "field": "key_metrics.2", "text": "300+ playbooks; 10,000+ invocations per week", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports", "locator": "Preserved content.md, lines 10" } ], "reported_by": "DoorDash", "metric_scope": "Unique playbooks and weekly invocations on Flux" }, { "id": "doordash-flux--lessons-learned-0", "approach_id": "doordash-flux", "field": "lessons_learned.0", "text": "Start narrow to earn trust; began with automated code review before CI triage, on-call, maintenance, ticket-driven dev", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--lessons-learned-1", "approach_id": "doordash-flux", "field": "lessons_learned.1", "text": "Make the work visible; public Slack threads drove adoption; private per-run channels did not build team habits", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--lessons-learned-2", "approach_id": "doordash-flux", "field": "lessons_learned.2", "text": "Playbooks need enablement; workshops and hackathons turn repeated operational work into reusable playbooks", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "doordash-flux--lessons-learned-3", "approach_id": "doordash-flux", "field": "lessons_learned.3", "text": "Earn complexity by exhausting simpler primitives first; keep swarms at the research frontier until governance catches up", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" } ] }, { "id": "doordash-flux--lessons-learned-4", "approach_id": "doordash-flux", "field": "lessons_learned.4", "text": "Deterministic verification before probabilistic judgment; SQL linting and EXPLAIN before deeper validation; LLM-as-judge + DeepEval", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" } ] }, { "id": "doordash-flux--lessons-learned-5", "approach_id": "doordash-flux", "field": "lessons_learned.5", "text": "Log provenance so any answer traces back to source queries, documents, and inter-agent activity", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "doordash-flux-source-2", "relation": "supports" } ] }, { "id": "doordash-flux--operating-models-0", "approach_id": "doordash-flux", "field": "operating_models.0", "text": "Level 3 for engineering task → reviewed agent output; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The platform spans several workflows; the cited engineering examples retain human review of agent output.", "valid_at": "2025-11-11", "evidence": [ { "source_id": "doordash-flux-source-3", "relation": "supports" } ] }, { "id": "dropbox-nova--summary", "approach_id": "dropbox-nova", "field": "summary", "text": "An internal platform for coding agents: engineers launch parallel sessions and internal systems invoke agents inside automated SDLC workflows.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--headline-metric", "approach_id": "dropbox-nova", "field": "headline_metric", "text": "Dozens of agents can run in parallel from one runbook", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports", "locator": "Preserved content.md, lines 65–69" } ], "reported_by": "Dropbox", "metric_scope": "Migration-owner orchestration of dozens of agents from a shared runbook; qualitative capacity description" }, { "id": "dropbox-nova--architecture-sandbox", "approach_id": "dropbox-nova", "field": "architecture.sandbox", "text": "Isolated env with a codebase snapshot at a specific commit; full Dropbox monorepo via Bazel; hermetic remote execution + caching", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--architecture-harness", "approach_id": "dropbox-nova", "field": "architecture.harness", "text": "Validation loop (propose → validate → feed back) with continue_on_validation_failure and max_iterations (~5); branch management kept outside the agent", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--architecture-model", "approach_id": "dropbox-nova", "field": "architecture.model", "text": "Platform-agnostic; multiple coding agents behind one interface; swap models without rebuilding infra; prompt-eval tooling", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--architecture-interfaces", "approach_id": "dropbox-nova", "field": "architecture.interfaces", "text": "web, cli, api, slack", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--architecture-tool-access", "approach_id": "dropbox-nova", "field": "architecture.tool_access", "text": "Skills/plugins to gather evidence, read logs, inspect failures; MCP integrations; Bazel-aware selectivity tools", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--architecture-knowledge", "approach_id": "dropbox-nova", "field": "architecture.knowledge", "text": "Localized AGENTS.md per service; Dash (Dropbox context engineering); passing + failing test logs", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--architecture-credentials", "approach_id": "dropbox-nova", "field": "architecture.credentials", "text": "Operates within Dropbox's existing infra and validation paths; same auth/authz as engineers", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--architecture-context-mgmt", "approach_id": "dropbox-nova", "field": "architecture.context_mgmt", "text": "Session history (notes/logs) carried across retry attempts", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--primitives-0", "approach_id": "dropbox-nova", "field": "primitives.0", "text": "Bounded iteration with feedback on failure; deterministic systems control test execution", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--primitives-1", "approach_id": "dropbox-nova", "field": "primitives.1", "text": "Dropbox context-engineering system feeding agents across the SDLC", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--key-metrics-0", "approach_id": "dropbox-nova", "field": "key_metrics.0", "text": "Flaky-test remediation (Deflaker): 100+ validation runs", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports", "locator": "Preserved content.md, lines 58–62" } ], "reported_by": "Dropbox", "metric_scope": "CI validation runs per proposed Deflaker flaky-test fix", "measurement_method": "Run the test 100 or more times depending on its failure rate; retry capped at five fix attempts" }, { "id": "dropbox-nova--key-metrics-1", "approach_id": "dropbox-nova", "field": "key_metrics.1", "text": "Predecessor Goose-based migrator used across thousands of migration entries before workflows moved onto Nova", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports", "locator": "Preserved content.md, lines 65–69" } ], "reported_by": "Dropbox", "metric_scope": "Predecessor Goose-based migrator, before workflows moved onto Nova" }, { "id": "dropbox-nova--key-metrics-2", "approach_id": "dropbox-nova", "field": "key_metrics.2", "text": "Dozens of agents launchable from one runbook", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports", "locator": "Preserved content.md, lines 65–69" } ], "reported_by": "Dropbox", "metric_scope": "Migration-owner orchestration of dozens of agents from a shared runbook; qualitative capacity description" }, { "id": "dropbox-nova--lessons-learned-0", "approach_id": "dropbox-nova", "field": "lessons_learned.0", "text": "Platform value exceeds code generation; validation, guardrails, and context matter as much", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--lessons-learned-1", "approach_id": "dropbox-nova", "field": "lessons_learned.1", "text": "Context, validation, and guardrails reinforce each other to make background work trustworthy", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--lessons-learned-2", "approach_id": "dropbox-nova", "field": "lessons_learned.2", "text": "Not every step belongs in the agent loop; deterministic systems should control test execution and timing", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--lessons-learned-3", "approach_id": "dropbox-nova", "field": "lessons_learned.3", "text": "Integrate with existing engineering infrastructure rather than building separate AI-specific workflows", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "dropbox-nova--operating-models-0", "approach_id": "dropbox-nova", "field": "operating_models.0", "text": "Unclassified for agent-assisted SDLC workflow → accepted change; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The source documents human participation but does not locate one consistent attention boundary across Nova workflows.", "valid_at": "2026-05-22", "evidence": [ { "source_id": "dropbox-nova-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--summary", "approach_id": "flex-investigation-agent", "field": "summary", "text": "A Slack agent for HSA/FSA payment operations that traces a payment end-to-end and, when it finds a software bug, prepares a PR with a proposed fix.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--architecture-sandbox", "approach_id": "flex-investigation-agent", "field": "architecture.sandbox", "text": "unknown", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The preserved sources do not document an execution sandbox; unknown does not mean absent.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--architecture-harness", "approach_id": "flex-investigation-agent", "field": "architecture.harness", "text": "Investigation-to-fix loop; underlying runtime not documented", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--architecture-model", "approach_id": "flex-investigation-agent", "field": "architecture.model", "text": "Not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--architecture-interfaces", "approach_id": "flex-investigation-agent", "field": "architecture.interfaces", "text": "slack", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--architecture-tool-access", "approach_id": "flex-investigation-agent", "field": "architecture.tool_access", "text": "Traces a payment end-to-end across payment systems; can open a PR with a proposed fix when a bug is found", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--primitives-0", "approach_id": "flex-investigation-agent", "field": "primitives.0", "text": "Payment trace → root-cause hypothesis → proposed code fix as a PR", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--lessons-learned-0", "approach_id": "flex-investigation-agent", "field": "lessons_learned.0", "text": "Start where correctness is observable; payment investigation produces artifacts (a trace, a hypothesis, a diff) that can be checked", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--lessons-learned-1", "approach_id": "flex-investigation-agent", "field": "lessons_learned.1", "text": "An ops agent that can prepare a fix (not just a report) closes the loop from investigation to code", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "flex-investigation-agent--operating-models-0", "approach_id": "flex-investigation-agent", "field": "operating_models.0", "text": "Level 3 for payment investigation → proposed code fix; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source describes a trace, diagnosis, and proposed pull request returned for review.", "valid_at": "2026", "evidence": [ { "source_id": "flex-investigation-agent-source-1", "relation": "supports" } ] }, { "id": "github-qubot--summary", "approach_id": "github-qubot", "field": "summary", "text": "GitHub's internal data-analytics agent, powered by GitHub Copilot. Any GitHub employee can ask a question about the company data warehouse in plain language and get an answer within seconds.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "github-qubot-source-1", "relation": "supports" } ] }, { "id": "github-qubot--headline-metric", "approach_id": "github-qubot", "field": "headline_metric", "text": "Hundreds of users run thousands of queries; data questions in internal Slack channels dropped", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "GitHub reported the adoption figures in its own engineering blog without independent verification.", "valid_at": "2026-06", "evidence": [ { "source_id": "github-qubot-source-1", "relation": "supports", "locator": "Preserved content.md, lines 70" } ], "reported_by": "GitHub", "metric_scope": "Internal GitHub Qubot users and queries; no exact count or measurement window" }, { "id": "github-qubot--architecture-model", "approach_id": "github-qubot", "field": "architecture.model", "text": "Powered by GitHub Copilot", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "github-qubot-source-1", "relation": "supports" } ] }, { "id": "github-qubot--architecture-tool-access", "approach_id": "github-qubot", "field": "architecture.tool_access", "text": "Queries GitHub's data warehouse", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "github-qubot-source-1", "relation": "supports" } ] }, { "id": "github-qubot--key-metrics-0", "approach_id": "github-qubot", "field": "key_metrics.0", "text": "Hundreds of users run thousands of queries", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "GitHub reported the adoption figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "github-qubot-source-1", "relation": "supports", "locator": "Preserved content.md, lines 70" } ], "reported_by": "GitHub", "metric_scope": "Internal GitHub Qubot users and queries; no exact count or measurement window" }, { "id": "github-qubot--key-metrics-1", "approach_id": "github-qubot", "field": "key_metrics.1", "text": "Volume of data questions in internal data and analytics Slack channels dropped", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "Company reports a qualitative decrease without before-and-after counts.", "valid_at": null, "evidence": [ { "source_id": "github-qubot-source-1", "relation": "supports", "locator": "Preserved content.md, lines 70" } ], "metric_scope": "Questions in GitHub internal data and analytics Slack channels" }, { "id": "github-qubot--operating-models-0", "approach_id": "github-qubot", "field": "operating_models.0", "text": "Unclassified for data question → warehouse answer; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The public evidence does not document where human attention returns in the question-and-answer flow.", "valid_at": "2026-06", "evidence": [ { "source_id": "github-qubot-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--summary", "approach_id": "harvey-spectre", "field": "summary", "text": "Harvey's internal collaborative cloud agent platform; reacts to incidents, bug reports, and Slack messages and produces reviewable diffs, branches, and PRs.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--architecture-sandbox", "approach_id": "harvey-spectre", "field": "architecture.sandbox", "text": "Isolated ephemeral execution environments; durable runs", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--architecture-harness", "approach_id": "harvey-spectre", "field": "architecture.harness", "text": "Collaborative cloud agent platform with explicit boundaries around GitHub, Datadog, Linear, and other connected systems", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--architecture-model", "approach_id": "harvey-spectre", "field": "architecture.model", "text": "Not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--architecture-interfaces", "approach_id": "harvey-spectre", "field": "architecture.interfaces", "text": "slack, web, automation", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--architecture-tool-access", "approach_id": "harvey-spectre", "field": "architecture.tool_access", "text": "Explicit tool boundaries; reacts to incidents, bug reports, customer feedback, and Slack messages", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--architecture-credentials", "approach_id": "harvey-spectre", "field": "architecture.credentials", "text": "Explicit tool boundaries; outputs are reviewable", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--architecture-context-mgmt", "approach_id": "harvey-spectre", "field": "architecture.context_mgmt", "text": "Durable runs over disposable execution environments", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--primitives-0", "approach_id": "harvey-spectre", "field": "primitives.0", "text": "Long-lived run state over throwaway compute", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--primitives-1", "approach_id": "harvey-spectre", "field": "primitives.1", "text": "Outputs are summaries, diffs, branches, PRs; not silent actions", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--lessons-learned-0", "approach_id": "harvey-spectre", "field": "lessons_learned.0", "text": "Make outputs reviewable (diffs, branches, PRs) rather than letting the agent act silently", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--lessons-learned-1", "approach_id": "harvey-spectre", "field": "lessons_learned.1", "text": "Implementation speed shifts the bottleneck toward review, prioritization, and coordination", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "harvey-spectre--lessons-learned-2", "approach_id": "harvey-spectre", "field": "lessons_learned.2", "text": "Harvey keeps its product-agent and security-agent platforms on separate substrates because they have different trust boundaries", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "harvey-spectre-source-2", "relation": "supports" } ] }, { "id": "harvey-spectre--operating-models-0", "approach_id": "harvey-spectre", "field": "operating_models.0", "text": "Level 3 for incident or request → reviewable diff or pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source explicitly frames diffs, branches, and pull requests as reviewable outputs.", "valid_at": "2026", "evidence": [ { "source_id": "harvey-spectre-source-1", "relation": "supports" } ] }, { "id": "hubspot-sidekick--summary", "approach_id": "hubspot-sidekick", "field": "summary", "text": "HubSpot's internal AI code-review agent. Sidekick reviews every pull request and uses a multi-model Judge Agent to filter comments before posting. Its review implementation moved from Claude Code on Crucible Kubernetes workloads to Aviator, HubSpot's internal Java agent framework; the later report does not specify Aviator's execution isolation.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14, 24–45, 75–87" }, { "source_id": "hubspot-sidekick-source-2", "relation": "contextualizes", "locator": "Preserved content.md, lines 30–57 (earlier Crucible implementation)" } ] }, { "id": "hubspot-sidekick--headline-metric", "approach_id": "hubspot-sidekick", "field": "headline_metric", "text": "Reviews every pull request and cut engineer feedback time by 90%", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "HubSpot reported the figures in its own engineering blog without independent verification.", "valid_at": "2026-03", "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14–18" } ], "reported_by": "HubSpot", "metric_scope": "Time for engineers to receive code feedback from Sidekick; not overall PR completion time" }, { "id": "hubspot-sidekick--architecture-harness", "approach_id": "hubspot-sidekick", "field": "architecture.harness", "text": "Aviator, an internal Java agent framework; replaced the earlier Claude Code review implementation on Crucible", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24–45" } ] }, { "id": "hubspot-sidekick--architecture-sandbox", "approach_id": "hubspot-sidekick", "field": "architecture.sandbox", "text": "unknown", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "Current review runs on Aviator; its execution isolation is not specified. Crucible Kubernetes workloads describe the predecessor implementation.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports", "locator": "Preserved content.md, lines 39–45" } ] }, { "id": "hubspot-sidekick--architecture-tool-access", "approach_id": "hubspot-sidekick", "field": "architecture.tool_access", "text": "Aviator framework for precise tool control", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports" } ] }, { "id": "hubspot-sidekick--architecture-interfaces", "approach_id": "hubspot-sidekick", "field": "architecture.interfaces", "text": "github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports" } ] }, { "id": "hubspot-sidekick--key-metrics-0", "approach_id": "hubspot-sidekick", "field": "key_metrics.0", "text": "Reviews every pull request", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "HubSpot reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14" } ], "reported_by": "HubSpot", "metric_scope": "Pull-request coverage after the six-month rollout", "denominator": "HubSpot pull requests" }, { "id": "hubspot-sidekick--key-metrics-1", "approach_id": "hubspot-sidekick", "field": "key_metrics.1", "text": "Engineer feedback time cut by 90%", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "HubSpot reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14–18" } ], "reported_by": "HubSpot", "metric_scope": "Time for engineers to receive code feedback from Sidekick; not overall PR completion time" }, { "id": "hubspot-sidekick--key-metrics-2", "approach_id": "hubspot-sidekick", "field": "key_metrics.2", "text": "Over 80% thumbs-up reaction rate on review feedback during the preceding couple of months", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "HubSpot reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports", "locator": "Preserved content.md, lines 99–116" } ], "reported_by": "HubSpot", "metric_scope": "Developer emoji reactions on review comments during the preceding couple of months", "denominator": "Thumbs-up and thumbs-down reactions; not all developers or all reviews", "measurement_method": "Emoji reactions and replies on review comments" }, { "id": "hubspot-sidekick--operating-models-0", "approach_id": "hubspot-sidekick", "field": "operating_models.0", "text": "Level 3 for pull request → AI review comments; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The engineering blog describes engineers acting on Sidekick review comments, which locates human attention at work-product review.", "valid_at": "2026-03", "evidence": [ { "source_id": "hubspot-sidekick-source-1", "relation": "supports" } ] }, { "id": "linear-agent--summary", "approach_id": "linear-agent", "field": "summary", "text": "A native agent that synthesizes workspace context, triages, creates follow-up work, runs coding sessions, and executes scheduled/event-driven 'Loops'; used by Linear's own CX, Product, and Engineering teams.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-1", "relation": "supports" }, { "source_id": "linear-agent-source-2", "relation": "supports" }, { "source_id": "linear-agent-source-6", "relation": "supports" } ] }, { "id": "linear-agent--architecture-sandbox", "approach_id": "linear-agent", "field": "architecture.sandbox", "text": "unknown", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The preserved sources do not document an execution sandbox; unknown does not mean absent.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-1", "relation": "supports" } ] }, { "id": "linear-agent--architecture-harness", "approach_id": "linear-agent", "field": "architecture.harness", "text": "Separate planning agent (triage/issue creation) and coding agent (code generation); Code Intelligence for codebase knowledge; scheduled/event-driven 'Loops'", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-1", "relation": "supports" }, { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--architecture-model", "approach_id": "linear-agent", "field": "architecture.model", "text": "Codex was used for internal pull-request review; other model choices are not detailed in the preserved sources", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--architecture-interfaces", "approach_id": "linear-agent", "field": "architecture.interfaces", "text": "slack, intercom, linear, github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--architecture-tool-access", "approach_id": "linear-agent", "field": "architecture.tool_access", "text": "Triage Intelligence (auto-route, dedup, label); GitHub; testing Code Intelligence + custom MCP servers", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--architecture-knowledge", "approach_id": "linear-agent", "field": "architecture.knowledge", "text": "Semantic/vector search evolved into agentic context acquisition across the workspace; Datadog/Sentry customer context", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--architecture-credentials", "approach_id": "linear-agent", "field": "architecture.credentials", "text": "The Agent SDK gives agents explicit identities, scoped OAuth tokens, assignable/mentionable handles, and visible human delegation; issues stay assigned to a human; 'an agent cannot be held accountable'", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-3", "relation": "supports" } ] }, { "id": "linear-agent--primitives-0", "approach_id": "linear-agent", "field": "primitives.0", "text": "Auto-routes issues, flags duplicates, suggests labels", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--primitives-1", "approach_id": "linear-agent", "field": "primitives.1", "text": "Agents get identities, scoped team access, and visible delegation alongside humans", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-3", "relation": "supports" } ] }, { "id": "linear-agent--primitives-2", "approach_id": "linear-agent", "field": "primitives.2", "text": "Scheduled or event-driven agent runs that execute recurring work", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-1", "relation": "supports" } ] }, { "id": "linear-agent--lessons-learned-0", "approach_id": "linear-agent", "field": "lessons_learned.0", "text": "Keep the agent close to the source of work (Intercom, Slack, Linear); the best workflows live where work already happens", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--lessons-learned-1", "approach_id": "linear-agent", "field": "lessons_learned.1", "text": "Gradual autonomy; start by asking for suggestions, observe, add guidance, only automate once proven reliable", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--lessons-learned-2", "approach_id": "linear-agent", "field": "lessons_learned.2", "text": "Break work into small steps to keep coding agents focused and successful", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--lessons-learned-3", "approach_id": "linear-agent", "field": "lessons_learned.3", "text": "One Linear engineer reports that agent mistakes reveal possible failure modes during review", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--lessons-learned-4", "approach_id": "linear-agent", "field": "lessons_learned.4", "text": "Close the loop; auto-notify the customer when their request ships", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "linear-agent-source-2", "relation": "supports" } ] }, { "id": "linear-agent--operating-models-0", "approach_id": "linear-agent", "field": "operating_models.0", "text": "Level 3 for assigned coding work → agent-created change; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The source documents delegated coding output while a human remains accountable for the assigned issue.", "valid_at": "2026-08-11", "evidence": [ { "source_id": "linear-agent-source-1", "relation": "supports" } ] }, { "id": "microsoft-prassistant--summary", "approach_id": "microsoft-prassistant", "field": "summary", "text": "Microsoft's internal AI code-review agent, built by the Developer Division Data and AI team. When an engineer creates a pull request, PRAssistant joins as a reviewer and leaves comments like a human reviewer. It is a distinct internal build that predates and later informed GitHub Copilot Pull Request Reviews.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "microsoft-prassistant-source-1", "relation": "supports" } ] }, { "id": "microsoft-prassistant--headline-metric", "approach_id": "microsoft-prassistant", "field": "headline_metric", "text": "Supports more than 90% of Microsoft PRs, impacting over 600,000 pull requests per month", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Microsoft reported the figures in its own engineering blog without independent verification.", "valid_at": "2025-07", "evidence": [ { "source_id": "microsoft-prassistant-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10" } ], "reported_by": "Microsoft", "metric_scope": "PRs supported by the internal AI review assistant across Microsoft", "denominator": "Company pull requests for coverage share" }, { "id": "microsoft-prassistant--architecture-interfaces", "approach_id": "microsoft-prassistant", "field": "architecture.interfaces", "text": "github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "microsoft-prassistant-source-1", "relation": "supports" } ] }, { "id": "microsoft-prassistant--key-metrics-0", "approach_id": "microsoft-prassistant", "field": "key_metrics.0", "text": "More than 90% of pull requests across the company", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Microsoft reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "microsoft-prassistant-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10" } ], "reported_by": "Microsoft", "metric_scope": "PRs supported by the internal AI review assistant across Microsoft", "denominator": "Company pull requests for coverage share" }, { "id": "microsoft-prassistant--key-metrics-1", "approach_id": "microsoft-prassistant", "field": "key_metrics.1", "text": "More than 600,000 pull requests impacted per month", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Microsoft reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "microsoft-prassistant-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10" } ], "reported_by": "Microsoft", "metric_scope": "PRs supported by the internal AI review assistant across Microsoft", "denominator": "Company pull requests for coverage share" }, { "id": "microsoft-prassistant--key-metrics-2", "approach_id": "microsoft-prassistant", "field": "key_metrics.2", "text": "About 5,000 repositories in early onboarding", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Microsoft reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "microsoft-prassistant-source-1", "relation": "supports", "locator": "Preserved content.md, lines 35" } ], "reported_by": "Microsoft", "metric_scope": "Repositories onboarded in early AI code-review experiments", "measurement_method": "Early experiments and data science studies across onboarded repositories" }, { "id": "microsoft-prassistant--operating-models-0", "approach_id": "microsoft-prassistant", "field": "operating_models.0", "text": "Level 3 for pull request → AI review comments; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The engineering blog describes engineers acting on PRAssistant review comments, which locates human attention at work-product review.", "valid_at": "2025-07", "evidence": [ { "source_id": "microsoft-prassistant-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--summary", "approach_id": "monday-sphera-atlas-morphex", "field": "summary", "text": "An internal agent system on Amazon Bedrock where agents have identities, managers, scopes, and performance scores; Atlas ships features, Morphex ships PRs autonomously.", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--headline-metric", "approach_id": "monday-sphera-atlas-morphex", "field": "headline_metric", "text": "Morphex: 19 of 20 PRs merge without human review", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports", "locator": "Preserved content.md, lines 148–152" } ], "reported_by": "monday.com", "metric_scope": "Morphex PRs that merge automatically after CI and Guardrails pass", "denominator": "Morphex pull requests; sample size and period not supplied" }, { "id": "monday-sphera-atlas-morphex--architecture-sandbox", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.sandbox", "text": "Amazon EKS, one pod per active session; a remote sandbox tests each PR before review; EFS workspace mount", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--architecture-harness", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.harness", "text": "Claude Agent SDK behind a thin monday-agent-sdk wrapper (provider neutrality, cold-start optimization, custom harness opinions)", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--architecture-model", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.model", "text": "Amazon Bedrock; Application Inference Profiles for routing; cross-region failover; PrivateLink (traffic stays in VPC)", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--architecture-interfaces", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.interfaces", "text": "slack, monday, github", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--architecture-tool-access", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.tool_access", "text": "Triggers via Slack @mention, monday item assignment, or GitHub PR review → SNS → per-team SQS → consumers; monday MCP servers underpin the Guardrails", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--architecture-knowledge", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.knowledge", "text": "File-based memory: MEMORY.md (cross-session) + diary/YYYY-MM-DD.md; sessions/repos/secrets on EFS; durable records on S3", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--architecture-credentials", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.credentials", "text": "Per-session secrets in AWS Secrets Manager; same RBAC as humans; real Slack/GitHub/monday accounts", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--architecture-context-mgmt", "approach_id": "monday-sphera-atlas-morphex", "field": "architecture.context_mgmt", "text": "Live state in ElastiCache (sub-ms); monday boards (Builders CoWORK) as shared state for tasks, status, handoffs", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--primitives-0", "approach_id": "monday-sphera-atlas-morphex", "field": "primitives.0", "text": "Stable identity + assigned human manager + scope + performance score", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--primitives-1", "approach_id": "monday-sphera-atlas-morphex", "field": "primitives.1", "text": "Automated review against monday standards (metrics, feature flags, security)", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--primitives-2", "approach_id": "monday-sphera-atlas-morphex", "field": "primitives.2", "text": "MEMORY.md + daily diary instead of vector retrieval", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--primitives-3", "approach_id": "monday-sphera-atlas-morphex", "field": "primitives.3", "text": "monday boards as the shared-state layer for human/agent collaboration", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--key-metrics-0", "approach_id": "monday-sphera-atlas-morphex", "field": "key_metrics.0", "text": "Morphex: 19 of 20 PRs merge automatically without human review", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports", "locator": "Preserved content.md, lines 148–152" } ], "reported_by": "monday.com", "metric_scope": "Morphex PRs that merge automatically after CI and Guardrails pass", "denominator": "Morphex pull requests; sample size and period not supplied" }, { "id": "monday-sphera-atlas-morphex--key-metrics-1", "approach_id": "monday-sphera-atlas-morphex", "field": "key_metrics.1", "text": "90% of Builders use AI coding tools monthly; adoption nearly doubled year over year", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10, 16, 22" } ], "reported_by": "monday.com", "metric_scope": "Monthly AI coding-tool adoption among monday Builders, including engineers, PMs, analysts, and designers", "denominator": "Builders; not exclusively engineers" }, { "id": "monday-sphera-atlas-morphex--key-metrics-2", "approach_id": "monday-sphera-atlas-morphex", "field": "key_metrics.2", "text": "Per-engineer PR throughput increased by more than 50%", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10, 23" } ], "reported_by": "monday.com", "metric_scope": "Per-engineer pull-request throughput", "denominator": "Engineers; baseline period and cohort size not specified" }, { "id": "monday-sphera-atlas-morphex--key-metrics-3", "approach_id": "monday-sphera-atlas-morphex", "field": "key_metrics.3", "text": "Guardrails catches ~25% of agent PRs before human review; low single-digit revert rate", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports", "locator": "Preserved content.md, lines 154–158" } ], "reported_by": "monday.com", "metric_scope": "Recent cut of top PR-generating agents; Guardrails rejection before review and reverts among merged PRs", "denominator": "Agent PRs for Guardrails rejection; merged PRs for revert rate" }, { "id": "monday-sphera-atlas-morphex--lessons-learned-0", "approach_id": "monday-sphera-atlas-morphex", "field": "lessons_learned.0", "text": "Evals from day one; should have been day one, not month nine", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--lessons-learned-1", "approach_id": "monday-sphera-atlas-morphex", "field": "lessons_learned.1", "text": "Skip the vector store; file-based memory (MEMORY.md) was the right answer", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--lessons-learned-2", "approach_id": "monday-sphera-atlas-morphex", "field": "lessons_learned.2", "text": "Remote-sandbox every PR before human review, with production-traffic replay", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--lessons-learned-3", "approach_id": "monday-sphera-atlas-morphex", "field": "lessons_learned.3", "text": "The existing auth/identity/deploy pipeline applies to agents; reuse it", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--lessons-learned-4", "approach_id": "monday-sphera-atlas-morphex", "field": "lessons_learned.4", "text": "AI engineering is building the feedback loops that let imperfect agents be trusted safely", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "monday-sphera-atlas-morphex--operating-models-0", "approach_id": "monday-sphera-atlas-morphex", "field": "operating_models.0", "text": "Level 4 for Atlas or Morphex feature task → tested and merged pull request; human attention boundary: outcome-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The secondary source reports predominantly automatic merges and automated guardrails, while humans manage tasks and outcomes.", "valid_at": "2026", "evidence": [ { "source_id": "monday-sphera-atlas-morphex-source-1", "relation": "supports" } ] }, { "id": "notion-custom-agents--summary", "approach_id": "notion-custom-agents", "field": "summary", "text": "Notion's Custom Agents platform, dogfooded internally across non-engineering teams such as IT ticketing, supply chain, procurement, and recruiting. By the end of alpha testing, Notion had more than 3,000 internal Custom Agents. Notion's own security team is one of the most active internal users. Notion rebuilt the agent harness three to five times as frontier models improved.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "notion-custom-agents-source-1", "relation": "supports" }, { "source_id": "notion-custom-agents-source-2", "relation": "supports" } ] }, { "id": "notion-custom-agents--headline-metric", "approach_id": "notion-custom-agents", "field": "headline_metric", "text": "More than 3,000 internal Custom Agents by end of alpha testing", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Notion reported the agent count in its own engineering blog without independent verification.", "valid_at": "2026-04", "evidence": [ { "source_id": "notion-custom-agents-source-2", "relation": "supports", "locator": "Preserved content.md, lines 51" } ], "reported_by": "Notion", "metric_scope": "Internal Notion Custom Agents at the end of alpha; excludes the separate customer alpha count" }, { "id": "notion-custom-agents--key-metrics-0", "approach_id": "notion-custom-agents", "field": "key_metrics.0", "text": "More than 3,000 internal Custom Agents by end of alpha testing", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Notion reported the agent count in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "notion-custom-agents-source-2", "relation": "supports", "locator": "Preserved content.md, lines 51" } ], "reported_by": "Notion", "metric_scope": "Internal Notion Custom Agents at the end of alpha; excludes the separate customer alpha count" }, { "id": "notion-custom-agents--key-metrics-1", "approach_id": "notion-custom-agents", "field": "key_metrics.1", "text": "Notion's security team is one of the most active internal users", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "Qualitative first-party description; no comparative activity count is supplied.", "valid_at": null, "evidence": [ { "source_id": "notion-custom-agents-source-2", "relation": "supports", "locator": "Preserved content.md, lines 57" } ], "metric_scope": "Notion security team internal Custom Agent use" }, { "id": "notion-custom-agents--key-metrics-2", "approach_id": "notion-custom-agents", "field": "key_metrics.2", "text": "Agent harness rebuilt three to five times as models improved", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Participants give differing approximate rebuild counts for harness, framework, and feature; not a precise engineering inventory.", "valid_at": null, "evidence": [ { "source_id": "notion-custom-agents-source-1", "relation": "supports", "locator": "Preserved content.md, lines 115, 275–277, 819" } ], "reported_by": "Notion", "metric_scope": "Notion agent/framework rebuilds recalled by participants; estimates range from three to five" }, { "id": "notion-custom-agents--operating-models-0", "approach_id": "notion-custom-agents", "field": "operating_models.0", "text": "Unclassified for cross-team internal tasks → Custom Agents output; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "Custom Agents spans many teams and workflows, so no single human-attention boundary applies.", "valid_at": "2026-04", "evidence": [ { "source_id": "notion-custom-agents-source-1", "relation": "supports" } ] }, { "id": "plaid-ai-annotator--summary", "approach_id": "plaid-ai-annotator", "field": "summary", "text": "Plaid's internal labeling agent for its own model training. AI Annotator automates large-scale labeling of anonymized transaction data, with human oversight on the labeled output. Plaid reports greater than 95% human alignment at a lower cost and time than manual labeling.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-ai-annotator-source-1", "relation": "supports" } ] }, { "id": "plaid-ai-annotator--headline-metric", "approach_id": "plaid-ai-annotator", "field": "headline_metric", "text": "Greater than 95% human alignment at lower cost and time than manual labeling", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figure in its own blog without independent verification.", "valid_at": "2025-06", "evidence": [ { "source_id": "plaid-ai-annotator-source-1", "relation": "supports", "locator": "Preserved content.md, lines 16–20" } ], "reported_by": "Plaid", "metric_scope": "Transaction labels generated by AI Annotator in early use", "denominator": "Labels compared with human judgments; sample size unspecified" }, { "id": "plaid-ai-annotator--architecture-knowledge", "approach_id": "plaid-ai-annotator", "field": "architecture.knowledge", "text": "Anonymized Plaid transaction data", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-ai-annotator-source-1", "relation": "supports" } ] }, { "id": "plaid-ai-annotator--key-metrics-0", "approach_id": "plaid-ai-annotator", "field": "key_metrics.0", "text": "Greater than 95% human alignment with labeled data", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figure in its own blog.", "valid_at": null, "evidence": [ { "source_id": "plaid-ai-annotator-source-1", "relation": "supports", "locator": "Preserved content.md, lines 16–20" } ], "reported_by": "Plaid", "metric_scope": "Transaction labels generated by AI Annotator in early use", "denominator": "Labels compared with human judgments; sample size unspecified" }, { "id": "plaid-ai-annotator--key-metrics-1", "approach_id": "plaid-ai-annotator", "field": "key_metrics.1", "text": "Lower cost and time than manual labeling", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Qualitative company comparison; neither actual costs nor elapsed-time measurements are supplied.", "valid_at": null, "evidence": [ { "source_id": "plaid-ai-annotator-source-1", "relation": "supports", "locator": "Preserved content.md, lines 20" } ], "reported_by": "Plaid", "metric_scope": "AI transaction annotation cost and time relative to manual labeling" }, { "id": "plaid-ai-annotator--operating-models-0", "approach_id": "plaid-ai-annotator", "field": "operating_models.0", "text": "Level 3 for raw transactions → labeled training data; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The reported greater than 95% human alignment implies that people review the labeled output.", "valid_at": "2025-06", "evidence": [ { "source_id": "plaid-ai-annotator-source-1", "relation": "supports" } ] }, { "id": "plaid-fix-my-connection--summary", "approach_id": "plaid-fix-my-connection", "field": "summary", "text": "Plaid's internal agent for bank-integration reliability. Fix My Connection proactively detects bank-integration failures and generates repair scripts automatically. Plaid reports more than 2 million successful user-permissioned logins and a 90% reduction in the average time to fix a degradation.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-fix-my-connection-source-1", "relation": "supports" } ] }, { "id": "plaid-fix-my-connection--headline-metric", "approach_id": "plaid-fix-my-connection", "field": "headline_metric", "text": "More than 2 million successful logins and 90% faster average repair", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figures in its own blog without independent verification.", "valid_at": "2025-06", "evidence": [ { "source_id": "plaid-fix-my-connection-source-1", "relation": "supports", "locator": "Preserved content.md, lines 26–32" } ], "reported_by": "Plaid", "metric_scope": "Automated repair of bank connections and the successful user-permissioned logins enabled by those repairs" }, { "id": "plaid-fix-my-connection--architecture-tool-access", "approach_id": "plaid-fix-my-connection", "field": "architecture.tool_access", "text": "Plaid bank-integration infrastructure", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-fix-my-connection-source-1", "relation": "supports" } ] }, { "id": "plaid-fix-my-connection--key-metrics-0", "approach_id": "plaid-fix-my-connection", "field": "key_metrics.0", "text": "More than 2 million successful user-permissioned logins", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figure in its own blog.", "valid_at": null, "evidence": [ { "source_id": "plaid-fix-my-connection-source-1", "relation": "supports", "locator": "Preserved content.md, lines 26–32" } ], "reported_by": "Plaid", "metric_scope": "Automated repair of bank connections and the successful user-permissioned logins enabled by those repairs" }, { "id": "plaid-fix-my-connection--key-metrics-1", "approach_id": "plaid-fix-my-connection", "field": "key_metrics.1", "text": "Average time to fix a degradation reduced by 90%", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figure in its own blog.", "valid_at": null, "evidence": [ { "source_id": "plaid-fix-my-connection-source-1", "relation": "supports", "locator": "Preserved content.md, lines 26–32" } ], "reported_by": "Plaid", "metric_scope": "Automated repair of bank connections and the successful user-permissioned logins enabled by those repairs", "denominator": "Average degradation-repair time before automated repairs; no baseline duration provided" }, { "id": "plaid-fix-my-connection--operating-models-0", "approach_id": "plaid-fix-my-connection", "field": "operating_models.0", "text": "Level 4 for integration degradation → repaired connection; human attention boundary: outcome-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "Plaid measures success by outcomes such as successful logins rather than per-repair inspection.", "valid_at": "2025-06", "evidence": [ { "source_id": "plaid-fix-my-connection-source-1", "relation": "supports" } ] }, { "id": "plaid-internal-mcp-server--summary", "approach_id": "plaid-internal-mcp-server", "field": "summary", "text": "Plaid's central internal Model Context Protocol server. Plaid built it because third-party MCP servers could not reach its internal data. The server integrates more than 20 tools and several internal services such as Jira, application logs, and data schemas, behind Plaid's identity-aware proxy and centralized authorization. Plaid reports thousands of tool calls and dozens of agents built on the server. Separately, Claude Code and Cursor are used by more than 80% of Plaid engineers; server adoption is not quantified.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports", "locator": "Preserved content.md, lines 38–45, 65–89" } ] }, { "id": "plaid-internal-mcp-server--headline-metric", "approach_id": "plaid-internal-mcp-server", "field": "headline_metric", "text": "Dozens of agents rely on the internal MCP server; Claude Code and Cursor are used by over 80% of engineers", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figures in its own engineering blog without independent verification.", "valid_at": "2025", "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports", "locator": "Preserved content.md, lines 38, 89" } ], "reported_by": "Plaid", "metric_scope": "Claude Code and Cursor adoption among Plaid engineers; separate from internal MCP server adoption", "denominator": "Plaid engineers for the AI-client usage share" }, { "id": "plaid-internal-mcp-server--architecture-harness", "approach_id": "plaid-internal-mcp-server", "field": "architecture.harness", "text": "Central internal MCP server that fronts vendor AI clients", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports" } ] }, { "id": "plaid-internal-mcp-server--architecture-tool-access", "approach_id": "plaid-internal-mcp-server", "field": "architecture.tool_access", "text": "More than 20 tools and several internal services (Jira, logs, schemas)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports" } ] }, { "id": "plaid-internal-mcp-server--architecture-credentials", "approach_id": "plaid-internal-mcp-server", "field": "architecture.credentials", "text": "Behind Plaid's identity-aware proxy and centralized authorization", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports" } ] }, { "id": "plaid-internal-mcp-server--key-metrics-0", "approach_id": "plaid-internal-mcp-server", "field": "key_metrics.0", "text": "Claude Code and Cursor are used by over 80% of Plaid engineers; the source does not report internal MCP server adoption share", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports", "locator": "Preserved content.md, lines 38, 89" } ], "reported_by": "Plaid", "metric_scope": "Claude Code and Cursor adoption among Plaid engineers; separate from internal MCP server adoption", "denominator": "Plaid engineers for the AI-client usage share" }, { "id": "plaid-internal-mcp-server--key-metrics-1", "approach_id": "plaid-internal-mcp-server", "field": "key_metrics.1", "text": "Thousands of tool calls and dozens of agents built on it", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Plaid reported the figures in its own engineering blog.", "valid_at": null, "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports", "locator": "Preserved content.md, lines 89" } ], "reported_by": "Plaid", "metric_scope": "Tool calls and agents relying on the internal MCP server across engineering, product, and support" }, { "id": "plaid-internal-mcp-server--operating-models-0", "approach_id": "plaid-internal-mcp-server", "field": "operating_models.0", "text": "Unclassified for engineer request → internal tool access; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The MCP server is a tool-access layer, not a workflow with a single human-attention boundary.", "valid_at": "2025", "evidence": [ { "source_id": "plaid-internal-mcp-server-source-1", "relation": "supports" } ] }, { "id": "posthog-stamphog--summary", "approach_id": "posthog-stamphog", "field": "summary", "text": "A GitHub-label-triggered PR approval agent that applies fail-closed deterministic safety gates, asks an LLM to check for showstoppers, autonomously approves eligible changes, and refuses or escalates the rest.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-1", "relation": "supports", "locator": "Add a PR auto-stamper" }, { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, How it works" } ] }, { "id": "posthog-stamphog--headline-metric", "approach_id": "posthog-stamphog", "field": "headline_metric", "text": "Handled 1,600 PRs in the previous month, as reported on July 9, 2026; roughly one in three merged main-repository PRs received its final approval during the reported quarter", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "PostHog reports two different windows in its July 9, 2026 article; exact monthly and quarterly boundaries are not supplied.", "valid_at": "2026-07", "evidence": [ { "source_id": "posthog-stamphog-source-1", "relation": "supports", "locator": "Preserved content.md, lines 105, 118" } ], "reported_by": "PostHog", "metric_scope": "Monthly PRs handled autonomously and quarterly final approvals in the main repository", "denominator": "Merged main-repository PRs for the quarterly share; absolute handled PRs for the monthly count", "measurement_method": "Company-reported production usage" }, { "id": "posthog-stamphog--architecture-harness", "approach_id": "posthog-stamphog", "field": "architecture.harness", "text": "A GitHub Action invokes a Python pipeline that fetches and classifies a PR, applies hard gates, waits for in-flight reviewer bots, runs an LLM review, and posts a verdict", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, How it works and Architecture" } ] }, { "id": "posthog-stamphog--architecture-model", "approach_id": "posthog-stamphog", "field": "architecture.model", "text": "Claude through the Claude Agent SDK, with Read, Grep, and Glob tools", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, LLM Review" } ] }, { "id": "posthog-stamphog--architecture-interfaces", "approach_id": "posthog-stamphog", "field": "architecture.interfaces", "text": "github, ci", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Usage" } ] }, { "id": "posthog-stamphog--architecture-tool-access", "approach_id": "posthog-stamphog", "field": "architecture.tool_access", "text": "Reads the diff and repository files plus trusted review-state, discussion, ownership, and reviewer signals", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, LLM Review" } ] }, { "id": "posthog-stamphog--architecture-knowledge", "approach_id": "posthog-stamphog", "field": "architecture.knowledge", "text": "Repository-specific deny categories, size and risk tiers calibrated from prior human approvals, review guidance, and ownership data", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Tiers and Empirical basis" } ] }, { "id": "posthog-stamphog--architecture-credentials", "approach_id": "posthog-stamphog", "field": "architecture.credentials", "text": "A dedicated Anthropic organization secret and a StampHog GitHub App token whose approvals satisfy branch protection", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Usage and final verdict" } ] }, { "id": "posthog-stamphog--architecture-context-mgmt", "approach_id": "posthog-stamphog", "field": "architecture.context_mgmt", "text": "Each run emits a versioned JSON evidence bundle retained as a CI artifact for 30 days; a sticky GitHub comment carries non-approval verdict history and labels preserve retry state", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Usage and Evidence bundle" } ] }, { "id": "posthog-stamphog--primitives-0", "approach_id": "posthog-stamphog", "field": "primitives.0", "text": "Draft state, conflicts, requested changes, sensitive paths, size ceilings, and risk tiers can block AI approval; the LLM may tighten but never loosen a gate", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, How it works" } ] }, { "id": "posthog-stamphog--primitives-1", "approach_id": "posthog-stamphog", "field": "primitives.1", "text": "Eligible changes can be approved while risky, ambiguous, or insufficiently assured changes are refused or escalated to a suitable human reviewer", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-1", "relation": "supports", "locator": "Add a PR auto-stamper" }, { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, LLM Review" } ] }, { "id": "posthog-stamphog--primitives-2", "approach_id": "posthog-stamphog", "field": "primitives.2", "text": "Each run records PR metadata, classification, gate results, reviewer output, and the final verdict", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Evidence bundle" } ] }, { "id": "posthog-stamphog--key-metrics-0", "approach_id": "posthog-stamphog", "field": "key_metrics.0", "text": "1,600 PRs handled autonomously in the previous month, as reported on July 9, 2026", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "July 9 is the report date; the source says last month without exact measurement boundaries.", "valid_at": "2026-07-09", "evidence": [ { "source_id": "posthog-stamphog-source-1", "relation": "supports", "locator": "Preserved content.md, lines 118" } ], "reported_by": "PostHog", "metric_scope": "PRs handled autonomously in the previous month, as reported on July 9, 2026; exact boundaries unspecified", "measurement_method": "Company-reported production usage" }, { "id": "posthog-stamphog--key-metrics-1", "approach_id": "posthog-stamphog", "field": "key_metrics.1", "text": "Roughly one in three PRs merged into PostHog's main repository received StampHog's final approval during the reported quarter", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026-07", "evidence": [ { "source_id": "posthog-stamphog-source-1", "relation": "supports", "locator": "Preserved content.md, lines 105" } ], "reported_by": "PostHog", "metric_scope": "Main-repository merged PRs receiving StampHog final approval during the reported quarter", "denominator": "Merged PRs in the main repository", "measurement_method": "Company-reported production usage" }, { "id": "posthog-stamphog--key-metrics-2", "approach_id": "posthog-stamphog", "field": "key_metrics.2", "text": "20% of PRs approved by StampHog in the July 28, 2026 report, at approximately $300 per month in tokens", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026-07", "evidence": [ { "source_id": "posthog-stamphog-source-2", "relation": "supports", "locator": "Preserved content.md, lines 131" } ], "reported_by": "PostHog", "metric_scope": "PostHog PRs approved by StampHog and monthly token cost in the July 28, 2026 report", "denominator": "PostHog PRs in the report's scope", "measurement_method": "Company-reported production usage and token spend" }, { "id": "posthog-stamphog--lessons-learned-0", "approach_id": "posthog-stamphog", "field": "lessons_learned.0", "text": "Use deterministic controls for known risks and allow the LLM to make approval stricter, never more permissive", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source code explicitly implements and documents this safety invariant.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, How it works" } ] }, { "id": "posthog-stamphog--lessons-learned-1", "approach_id": "posthog-stamphog", "field": "lessons_learned.1", "text": "Calibrate thresholds and deny categories from repository history rather than treating small diffs as inherently safe", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source code documents calibration against historical approval outcomes.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Tiers and Empirical basis" } ] }, { "id": "posthog-stamphog--lessons-learned-2", "approach_id": "posthog-stamphog", "field": "lessons_learned.2", "text": "Fail closed and preserve retry state when dependencies, credentials, or concurrent reviewer bots are unavailable", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source code explicitly documents fail-closed and retry behavior.", "valid_at": null, "evidence": [ { "source_id": "posthog-stamphog-source-3", "relation": "supports", "locator": "commit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Usage" } ] }, { "id": "posthog-stamphog--operating-models-0", "approach_id": "posthog-stamphog", "field": "operating_models.0", "text": "Level 5 for eligible pull request → approval decision; human attention boundary: exception-only.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "Eligible pull requests are approved automatically; risky or ambiguous cases are refused or routed to a human.", "valid_at": "2026-07-09", "evidence": [ { "source_id": "posthog-stamphog-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--summary", "approach_id": "ramp-inspect", "field": "summary", "text": "A background coding agent that closes the loop on verifying its own work; runs tests, reviews telemetry, queries feature flags, visually verifies the frontend; now also monitoring production and proposing fixes; also a platform that hosts many internal agents.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" }, { "source_id": "ramp-inspect-source-4", "relation": "supports" }, { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "What is Inspect? section" } ] }, { "id": "ramp-inspect--headline-metric", "approach_id": "ramp-inspect", "field": "headline_metric", "text": "75% of Ramp's merged PRs raised by Inspect sessions (May 2026)", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2026-05", "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 58–68" } ], "reported_by": "Ramp", "metric_scope": "Ramp merged PRs raised by Inspect sessions by May 2026", "denominator": "Merged Ramp pull requests" }, { "id": "ramp-inspect--architecture-sandbox", "approach_id": "ramp-inspect", "field": "architecture.sandbox", "text": "Modal sandboxes; per-repo images rebuilt every 30 min from snapshots; warm-on-keystroke; a pool of warm sandboxes", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--architecture-harness", "approach_id": "ramp-inspect", "field": "architecture.harness", "text": "OpenCode (server-first) as the agent runtime; a plugin blocks writes until sync completes; expanded into production monitoring and self-maintenance; Inspect itself is built with React/Vite, Cloudflare Durable Objects, SQLite, and the Cloudflare Agents SDK", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--architecture-model", "approach_id": "ramp-inspect", "field": "architecture.model", "text": "All frontier models, MCPs, custom tools, skills", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--architecture-interfaces", "approach_id": "ramp-inspect", "field": "architecture.interfaces", "text": "slack, web, chrome-extension, github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--architecture-tool-access", "approach_id": "ramp-inspect", "field": "architecture.tool_access", "text": "Wired into Sentry, Datadog, LaunchDarkly, Braintrust, GitHub, Slack, Buildkite; monitors production, triages issues, proposes fixes; debugging queries a sanitized read-only production DB replica and Snowflake", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--architecture-knowledge", "approach_id": "ramp-inspect", "field": "architecture.knowledge", "text": "Skills that encode how Ramp ships; repo images with the full dev env (Vite, Postgres, Redis, RabbitMQ, Temporal, Chromium, VS Code Server)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--architecture-credentials", "approach_id": "ramp-inspect", "field": "architecture.credentials", "text": "GitHub auth per user; the sandbox pushes the branch, an API opens the PR with the user's token (no self-approval); production merges retain human review", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--primitives-0", "approach_id": "ramp-inspect", "field": "primitives.0", "text": "Pre-warmed full dev envs; fast cold start; effectively free to run", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--primitives-1", "approach_id": "ramp-inspect", "field": "primitives.1", "text": "Server-first agent with a typed SDK + plugin system; code is its own source of truth", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--primitives-2", "approach_id": "ramp-inspect", "field": "primitives.2", "text": "Any number of people in one session; each change attributed to its author", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--key-metrics-0", "approach_id": "ramp-inspect", "field": "key_metrics.0", "text": "Around 60% of Ramp PRs authored by Inspect by January 2026", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2026-01", "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 58–68" } ], "reported_by": "Ramp", "metric_scope": "Ramp PRs authored by Inspect by January 2026", "denominator": "Ramp pull requests in the source adoption history" }, { "id": "ramp-inspect--key-metrics-1", "approach_id": "ramp-inspect", "field": "key_metrics.1", "text": "75% of merged PRs raised by Inspect sessions (May 2026)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2026-05", "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 58–68" }, { "source_id": "ramp-inspect-source-2", "relation": "supports", "locator": "Preserved content.md, lines 10" } ], "reported_by": "Ramp", "metric_scope": "Ramp merged PRs raised by Inspect sessions by May 2026", "denominator": "Merged Ramp pull requests" }, { "id": "ramp-inspect--key-metrics-2", "approach_id": "ramp-inspect", "field": "key_metrics.2", "text": "Around 30% of merged frontend and backend PRs in the earlier first-party report, after a couple of months of adoption", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24" } ], "reported_by": "Ramp", "metric_scope": "Merged PRs in Ramp frontend and backend repositories after the first couple of months of adoption", "denominator": "Merged PRs in frontend and backend repositories" }, { "id": "ramp-inspect--key-metrics-3", "approach_id": "ramp-inspect", "field": "key_metrics.3", "text": "Around 90% of PRs merged into the Inspect repository come from Inspect sessions in the later interview report", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 96–102" } ], "reported_by": "Ramp", "metric_scope": "PRs merged into the Inspect repository", "denominator": "Merged PRs in the Inspect repository" }, { "id": "ramp-inspect--key-metrics-4", "approach_id": "ramp-inspect", "field": "key_metrics.4", "text": "One million total Inspect sessions crossed in July 2026", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": "2026-07", "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 66–72" } ], "reported_by": "Ramp", "metric_scope": "Cumulative Inspect sessions crossing the million mark in July 2026" }, { "id": "ramp-inspect--key-metrics-5", "approach_id": "ramp-inspect", "field": "key_metrics.5", "text": "Under 5 seconds to spin up a fully provisioned remote dev environment", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 100" } ], "reported_by": "Ramp", "metric_scope": "Provisioning a fully configured remote development environment" }, { "id": "ramp-inspect--key-metrics-6", "approach_id": "ramp-inspect", "field": "key_metrics.6", "text": "Inspect underpins internal agents including ReviewBuddy, Oncall Assistant, Testo, Ramp Research, Voice of the Customer, and error automations", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 134–141" } ], "metric_scope": "Examples of internal agents built on Inspect" }, { "id": "ramp-inspect--key-metrics-7", "approach_id": "ramp-inspect", "field": "key_metrics.7", "text": "150+ engineers contributed to the Inspect codebase in the later interview report", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 102" } ], "reported_by": "Ramp", "metric_scope": "Engineers who contributed to the Inspect codebase in the later interview report" }, { "id": "ramp-inspect--key-metrics-8", "approach_id": "ramp-inspect", "field": "key_metrics.8", "text": "5.5-person Inspect team (four engineers, a director, and a part-time PM)", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 101" } ], "reported_by": "Ramp", "metric_scope": "Inspect team staffing: four engineers, one director, and a part-time PM" }, { "id": "ramp-inspect--key-metrics-9", "approach_id": "ramp-inspect", "field": "key_metrics.9", "text": "More than 80% of Inspect is written in Inspect sessions", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked participant or independent source reports the claim.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Preserved content.md, lines 111" } ], "reported_by": "Ramp", "metric_scope": "Inspect code written in Inspect sessions; distinct from merged PR share", "denominator": "Inspect code; the source does not define a code-volume counting method" }, { "id": "ramp-inspect--key-metrics-10", "approach_id": "ramp-inspect", "field": "key_metrics.10", "text": "Design goal: session speed should be limited only by model-provider time-to-first-token", "kind": "opinion", "provenance": "reported", "confidence": "medium", "confidence_reason": "The source says session speed should only be limited by model time-to-first-token; this is a design goal, not a measured result.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports", "locator": "Preserved content.md, lines 22" } ], "metric_scope": "Target session startup speed, excluding precompleted cloning and installation" }, { "id": "ramp-inspect--lessons-learned-0", "approach_id": "ramp-inspect", "field": "lessons_learned.0", "text": "Own the tooling; it only has to work on your code, which lets you build something more powerful than off-the-shelf", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--lessons-learned-1", "approach_id": "ramp-inspect", "field": "lessons_learned.1", "text": "Work in public spaces to create virality loops; let the product do the talking, don't mandate", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--lessons-learned-2", "approach_id": "ramp-inspect", "field": "lessons_learned.2", "text": "Ramp argues that a fast background agent can add remote resources and concurrency to the same model", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--lessons-learned-3", "approach_id": "ramp-inspect", "field": "lessons_learned.3", "text": "Move as much as possible into the image-build step so users never wait on setup", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "ramp-inspect--lessons-learned-4", "approach_id": "ramp-inspect", "field": "lessons_learned.4", "text": "The v1 Chrome extension saw little adoption; the pivot to a centrally configured remote dev environment with a coding agent on top drove adoption", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The source reports that v1 saw little adoption and that the November 2025 pivot to a remote dev environment preceded rapid adoption.", "valid_at": null, "evidence": [ { "source_id": "ramp-inspect-source-5", "relation": "supports", "locator": "Why build your own background coding agent? section" } ] }, { "id": "ramp-inspect--operating-models-0", "approach_id": "ramp-inspect", "field": "operating_models.0", "text": "Level 3 for Inspect coding task → reviewed production merge; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source explicitly states that Inspect cannot self-approve and production merges retain human review.", "valid_at": "2026-05", "evidence": [ { "source_id": "ramp-inspect-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--summary", "approach_id": "replit-manager-agent", "field": "summary", "text": "An internal agent-of-agents stack where every employee gets a manager agent that spawns multiple agents for verifiable work and escalates judgment to humans.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--headline-metric", "approach_id": "replit-manager-agent", "field": "headline_metric", "text": "2.9x code output for a consistent author cohort; review latency, PR reversions, and incident trends reported flat", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10, 38–56" } ], "reported_by": "Replit", "metric_scope": "Code output for a consistent author cohort across early January to late June; separate from company-wide hiring effects", "denominator": "Same cohort of authors before and after", "measurement_method": "Comparison of contributed code for a consistent author cohort; raw company-wide lines of code increased 5.8x" }, { "id": "replit-manager-agent--architecture-sandbox", "approach_id": "replit-manager-agent", "field": "architecture.sandbox", "text": "microVMs and remote filesystems behind access policies, token proxies, audit logging, and a ZeroTrust network", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports", "locator": "Preserved content.md, lines 34" } ] }, { "id": "replit-manager-agent--architecture-harness", "approach_id": "replit-manager-agent", "field": "architecture.harness", "text": "Fleet/loop orchestration: a manager agent launches parallel agents for verifiable work and escalates judgment", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--architecture-model", "approach_id": "replit-manager-agent", "field": "architecture.model", "text": "Not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--architecture-interfaces", "approach_id": "replit-manager-agent", "field": "architecture.interfaces", "text": "slack", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--architecture-tool-access", "approach_id": "replit-manager-agent", "field": "architecture.tool_access", "text": "Investigates incidents, reviews PRs, answers questions, analyzes company data, triages support, researches sales accounts, improves Replit Agent itself", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--architecture-context-mgmt", "approach_id": "replit-manager-agent", "field": "architecture.context_mgmt", "text": "Manager agent coordinates parallel sub-agents and routes results", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--primitives-0", "approach_id": "replit-manager-agent", "field": "primitives.0", "text": "One human gives an objective; the manager spawns parallel agents for verifiable work and escalates judgment", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--key-metrics-0", "approach_id": "replit-manager-agent", "field": "key_metrics.0", "text": "2.9x code output for a consistent author cohort from early January to late June", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10, 38–56" } ], "reported_by": "Replit", "metric_scope": "Code output for a consistent author cohort across early January to late June; separate from company-wide hiring effects", "denominator": "Same cohort of authors before and after", "measurement_method": "Comparison of contributed code for a consistent author cohort; raw company-wide lines of code increased 5.8x" }, { "id": "replit-manager-agent--key-metrics-1", "approach_id": "replit-manager-agent", "field": "key_metrics.1", "text": "No corresponding deterioration in review/reversion/incident metrics", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports", "locator": "Preserved content.md, lines 50–56" } ], "reported_by": "Replit", "metric_scope": "Company code review latency, PR reversion rates, and incidents opened during increased code output", "measurement_method": "Company comparison of review latency, PR reversion rates, and incident trends" }, { "id": "replit-manager-agent--lessons-learned-0", "approach_id": "replit-manager-agent", "field": "lessons_learned.0", "text": "Give every employee a manager agent that spawns sub-agents; verifiable work parallelizes, judgment escalates to humans", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--lessons-learned-1", "approach_id": "replit-manager-agent", "field": "lessons_learned.1", "text": "Track outcome metrics (reverts, incidents), not activity; output can scale without quality regressions", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "replit-manager-agent--operating-models-0", "approach_id": "replit-manager-agent", "field": "operating_models.0", "text": "Level 3 for objective → verifiable multi-agent work product; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source describes autonomous parallel execution followed by human judgment on the resulting work.", "valid_at": "2026", "evidence": [ { "source_id": "replit-manager-agent-source-1", "relation": "supports" } ] }, { "id": "retool-retoolgpt--summary", "approach_id": "retool-retoolgpt", "field": "summary", "text": "Retool's internal assistant, built as a version of ChatGPT with access to Retool's internal Confluence documents, Retool documentation, and Linear tickets. The team deployed it organization-wide in a read-only environment so the whole team could use it.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "retool-retoolgpt-source-1", "relation": "supports" }, { "source_id": "retool-retoolgpt-source-2", "relation": "supports" } ] }, { "id": "retool-retoolgpt--architecture-model", "approach_id": "retool-retoolgpt", "field": "architecture.model", "text": "Built on ChatGPT", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "retool-retoolgpt-source-1", "relation": "supports" } ] }, { "id": "retool-retoolgpt--architecture-knowledge", "approach_id": "retool-retoolgpt", "field": "architecture.knowledge", "text": "Retool Confluence documents, Retool documentation, and Linear tickets", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "retool-retoolgpt-source-1", "relation": "supports" } ] }, { "id": "retool-retoolgpt--operating-models-0", "approach_id": "retool-retoolgpt", "field": "operating_models.0", "text": "Unclassified for internal question → sourced answer; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The public evidence does not document where human attention returns in the question-and-answer flow.", "valid_at": "2025-08", "evidence": [ { "source_id": "retool-retoolgpt-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--summary", "approach_id": "salesforce-slackbot", "field": "summary", "text": "Salesforce was 'customer zero' for the rebuilt Slackbot; an employee agent that finds company context, drafts work, and connects Slack context with Salesforce data; now also an external product.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" }, { "source_id": "salesforce-slackbot-source-2", "relation": "supports" } ] }, { "id": "salesforce-slackbot--architecture-harness", "approach_id": "salesforce-slackbot", "field": "architecture.harness", "text": "An employee agent intended as a front door to other agents", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--architecture-model", "approach_id": "salesforce-slackbot", "field": "architecture.model", "text": "Not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--architecture-interfaces", "approach_id": "salesforce-slackbot", "field": "architecture.interfaces", "text": "slack", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--architecture-tool-access", "approach_id": "salesforce-slackbot", "field": "architecture.tool_access", "text": "Finds company context, drafts work, manages meetings, connects Slack context with Salesforce data", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--architecture-knowledge", "approach_id": "salesforce-slackbot", "field": "architecture.knowledge", "text": "Slack context joined with Salesforce CRM data", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--architecture-credentials", "approach_id": "salesforce-slackbot", "field": "architecture.credentials", "text": "Permission-aware by construction; sees what the employee can see, respects roles and access controls", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--primitives-0", "approach_id": "salesforce-slackbot", "field": "primitives.0", "text": "Agent visibility is bounded by the invoking employee's permissions", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--lessons-learned-0", "approach_id": "salesforce-slackbot", "field": "lessons_learned.0", "text": "Dogfood internally first ('customer zero') before shipping externally", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--lessons-learned-1", "approach_id": "salesforce-slackbot", "field": "lessons_learned.1", "text": "Make permission-awareness a construction property, not a prompt instruction; the agent sees only what the employee can see", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "salesforce-slackbot--operating-models-0", "approach_id": "salesforce-slackbot", "field": "operating_models.0", "text": "Level 3 for employee request → drafted work; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The source describes an employee agent that prepares drafts and contextual work for human use.", "valid_at": "2026-01-14", "evidence": [ { "source_id": "salesforce-slackbot-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--summary", "approach_id": "sentry-junior", "field": "summary", "text": "An open-source Slack agent built at Sentry that acts like an intern; takes tasks, retrieves context across many company systems, and is steered and reviewed by humans. Its CEO argues one general-purpose agent beat several vendor-specific bots.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" }, { "source_id": "sentry-junior-source-2", "relation": "supports" }, { "source_id": "sentry-junior-source-3", "relation": "contextualizes" } ] }, { "id": "sentry-junior--headline-metric", "approach_id": "sentry-junior", "field": "headline_metric", "text": "Open-source (Apache-2.0) Slack agent (~100k lines of TS) used internally at Sentry", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10–24, 34–60" }, { "source_id": "sentry-junior-source-4", "relation": "supports", "locator": "Preserved content.md, lines 412–414" } ], "reported_by": "Sentry", "metric_scope": "Junior codebase size and license in the author report" }, { "id": "sentry-junior--architecture-sandbox", "approach_id": "sentry-junior", "field": "architecture.sandbox", "text": "Vercel serverless functions; Vercel agent-browser sandbox with an on-path proxy for traffic interception; ephemeral containers", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--architecture-harness", "approach_id": "sentry-junior", "field": "architecture.harness", "text": "Custom harness on Pi's SDK; a task broker over Vercel Queues with an inbox -> worker-claim -> interrupt/resume pattern to survive serverless timeouts; 'skills-as-runbooks'", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--architecture-model", "approach_id": "sentry-junior", "field": "architecture.model", "text": "Claude Sonnet (faster); swappable (Opus as a more expensive option)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--architecture-interfaces", "approach_id": "sentry-junior", "field": "architecture.interfaces", "text": "slack, web, github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--architecture-tool-access", "approach_id": "sentry-junior", "field": "architecture.tool_access", "text": "Progressive discovery via MCP; by default Junior connects to no provider until the agent requests a tool lookup; plugins connect Sentry, GitHub, Linear, Notion", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--architecture-knowledge", "approach_id": "sentry-junior", "field": "architecture.knowledge", "text": "Conversation transcripts persisted in Redis; repo search to trace code paths; skill docs (TELEMETRY.md, SOUL.md)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--architecture-credentials", "approach_id": "sentry-junior", "field": "architecture.credentials", "text": "On-path proxy injection; the model never sees the token because it is not in the sandbox; plugins declare OAuth flows and credential domains; GitHub distinguishes read vs. write", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--architecture-context-mgmt", "approach_id": "sentry-junior", "field": "architecture.context_mgmt", "text": "Incremental transcript updates in Redis; resource subscriptions to GitHub PR events for follow-ups", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--primitives-0", "approach_id": "sentry-junior", "field": "primitives.0", "text": "searchMcpTools loads tools on demand instead of dumping every schema into the prompt", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--primitives-1", "approach_id": "sentry-junior", "field": "primitives.1", "text": "Credentials injected host-side; the sandbox/model never touches a secret", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--primitives-2", "approach_id": "sentry-junior", "field": "primitives.2", "text": "Survives serverless timeouts via a queue + claim model", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--key-metrics-0", "approach_id": "sentry-junior", "field": "key_metrics.0", "text": "Around 100,000 lines of TypeScript excluding tests, evals, docs, and lockfiles", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports", "locator": "Preserved content.md, lines 22" } ], "reported_by": "Sentry", "metric_scope": "TypeScript lines in Junior excluding tests, evals, documentation, and lockfiles" }, { "id": "sentry-junior--key-metrics-1", "approach_id": "sentry-junior", "field": "key_metrics.1", "text": "4 months from start to writeup", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports", "locator": "Preserved content.md, lines 10–22" } ], "reported_by": "Sentry", "metric_scope": "Author-reported development and iteration time before the writeup" }, { "id": "sentry-junior--lessons-learned-0", "approach_id": "sentry-junior", "field": "lessons_learned.0", "text": "One general-purpose agent connected to many company systems beats several vendor-specific bots", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--lessons-learned-1", "approach_id": "sentry-junior", "field": "lessons_learned.1", "text": "Skills-as-runbooks encode operational knowledge the agent can follow", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--lessons-learned-2", "approach_id": "sentry-junior", "field": "lessons_learned.2", "text": "Stateless compute fights you; serverless functions time out and disappear; model the agent around interrupt/resume", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--lessons-learned-3", "approach_id": "sentry-junior", "field": "lessons_learned.3", "text": "Unit tests are the wrong yardstick for agents; invest in evals and integration tests instead", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--lessons-learned-4", "approach_id": "sentry-junior", "field": "lessons_learned.4", "text": "Writes need per-user authorization, not blanket trust", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "sentry-junior--operating-models-0", "approach_id": "sentry-junior", "field": "operating_models.0", "text": "Level 2 for assigned task → human-steered and reviewed output; human attention boundary: continuous-steering.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source explicitly describes humans steering and reviewing the agent throughout its work.", "valid_at": "2026", "evidence": [ { "source_id": "sentry-junior-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--summary", "approach_id": "shopify-internal-agents", "field": "summary", "text": "The Aquifer agent platform (session/harness/sandbox split) powers River, a Slack-native coding agent, plus research, migration, and app-security agents.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--headline-metric", "approach_id": "shopify-internal-agents", "field": "headline_metric", "text": "1 in 8 merged PRs company-wide coauthored by River", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports", "locator": "Preserved content.md, lines 12" } ], "reported_by": "Shopify", "metric_scope": "Merged River-coauthored PRs across Shopify", "denominator": "All Shopify merged pull requests" }, { "id": "shopify-internal-agents--architecture-sandbox", "approach_id": "shopify-internal-agents", "field": "architecture.sandbox", "text": "An execution environment (filesystem, shell, repo, build/test) separated from the harness; Shopify credits the brain-and-hands framing to Anthropic", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--architecture-harness", "approach_id": "shopify-internal-agents", "field": "architecture.harness", "text": "Session (durable; Postgres append-only event log) + Harness (cheap agent loop) + Cell (ephemeral Go runtime); cells die, sessions persist", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--architecture-model", "approach_id": "shopify-internal-agents", "field": "architecture.model", "text": "The design lets Shopify change the model without changing the sandbox", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--architecture-interfaces", "approach_id": "shopify-internal-agents", "field": "architecture.interfaces", "text": "slack, github", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--architecture-tool-access", "approach_id": "shopify-internal-agents", "field": "architecture.tool_access", "text": "Repo + tests + data warehouse + production traces + PR creation; a gateway credentials proxy", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--architecture-knowledge", "approach_id": "shopify-internal-agents", "field": "architecture.knowledge", "text": "Monorepo 'World' (code + skills + conventions + intent docs + runbooks + AGENTS.md); Nix reproducible envs; Slack-transcript corpus mining", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--architecture-credentials", "approach_id": "shopify-internal-agents", "field": "architecture.credentials", "text": "Gateway credentials proxy; per-profile sandbox policies; Shopify SSO", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--architecture-context-mgmt", "approach_id": "shopify-internal-agents", "field": "architecture.context_mgmt", "text": "Skills loaded on-demand as files, updatable per session; session survival across cell/sandbox/machine death", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--primitives-0", "approach_id": "shopify-internal-agents", "field": "primitives.0", "text": "Durable identity, disposable loop, isolated execution; swap any layer independently", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--primitives-1", "approach_id": "shopify-internal-agents", "field": "primitives.1", "text": "Slack-native coding agent in public channels only; visibility drives adoption", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--primitives-2", "approach_id": "shopify-internal-agents", "field": "primitives.2", "text": "Written-down knowledge mined from successful patterns and public transcripts", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--key-metrics-0", "approach_id": "shopify-internal-agents", "field": "key_metrics.0", "text": "River: 59,918 sessions / 30 days across 5,170 Slack channels", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports", "locator": "Preserved content.md, lines 83" } ], "reported_by": "Shopify", "metric_scope": "River sessions and distinct Slack channels in a recent 30-day period", "measurement_method": "river_sessions domain table, written by River every session" }, { "id": "shopify-internal-agents--key-metrics-1", "approach_id": "shopify-internal-agents", "field": "key_metrics.1", "text": "3,536 River-coauthored PRs merged; 1 in 8 merged PRs company-wide", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports", "locator": "Preserved content.md, lines 12, 83" } ], "reported_by": "Shopify", "metric_scope": "River-coauthored merged PRs in the recent 30-day period; company-wide PR share", "denominator": "All Shopify merged pull requests for the one-in-eight share", "measurement_method": "river_sessions domain table for reported session-linked counts" }, { "id": "shopify-internal-agents--key-metrics-2", "approach_id": "shopify-internal-agents", "field": "key_metrics.2", "text": "Median session 19 min; median 50 tool calls/session", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports", "locator": "Preserved content.md, lines 65" } ], "reported_by": "Shopify", "metric_scope": "Median River session duration and tool calls per session" }, { "id": "shopify-internal-agents--lessons-learned-0", "approach_id": "shopify-internal-agents", "field": "lessons_learned.0", "text": "Agent-friendly is human-friendly; monorepo, reproducible envs, written skills, and fast CI help both", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--lessons-learned-1", "approach_id": "shopify-internal-agents", "field": "lessons_learned.1", "text": "Local agents have a ceiling; private windows mean only the person at the keyboard learns anything", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--lessons-learned-2", "approach_id": "shopify-internal-agents", "field": "lessons_learned.2", "text": "Session survival is critical; cells die, sandboxes die, machines die; the conversation doesn't", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--lessons-learned-3", "approach_id": "shopify-internal-agents", "field": "lessons_learned.3", "text": "Treat agents as profiles, not platforms; a new agent is a new bundle on the same substrate", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "shopify-internal-agents--operating-models-0", "approach_id": "shopify-internal-agents", "field": "operating_models.0", "text": "Level 3 for River coding request → reviewed pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The source documents pull-request creation but does not establish outcome-only supervision.", "valid_at": "2026", "evidence": [ { "source_id": "shopify-internal-agents-source-1", "relation": "supports" } ] }, { "id": "sierra-pinecone--summary", "approach_id": "sierra-pinecone", "field": "summary", "text": "One company-wide agent that collapsed separate support, analytics, engineering, and sales agents into a single runtime with an MCP Gateway to 45 systems.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" }, { "source_id": "sierra-pinecone-source-2", "relation": "supports" } ] }, { "id": "sierra-pinecone--headline-metric", "approach_id": "sierra-pinecone", "field": "headline_metric", "text": "More than 75,000 sessions created by 600 people in the month preceding the report", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports", "locator": "Preserved content.md, lines 134–139" } ], "reported_by": "Sierra", "metric_scope": "Pinecone users and sessions in the month preceding the report; calendar month unspecified" }, { "id": "sierra-pinecone--architecture-sandbox", "approach_id": "sierra-pinecone", "field": "architecture.sandbox", "text": "Agency layer reconciles recoverable Kubernetes runners; conversation/events/checkpoints stay durable separately", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--architecture-harness", "approach_id": "sierra-pinecone", "field": "architecture.harness", "text": "App server + Agency + runners; intent-based model/environment routing (Claude Code + Codex)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" }, { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--architecture-model", "approach_id": "sierra-pinecone", "field": "architecture.model", "text": "Claude Code + Codex; routes by intent (planning, coding, prose)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" } ] }, { "id": "sierra-pinecone--architecture-interfaces", "approach_id": "sierra-pinecone", "field": "architecture.interfaces", "text": "slack, web, linear, mobile", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" } ] }, { "id": "sierra-pinecone--architecture-tool-access", "approach_id": "sierra-pinecone", "field": "architecture.tool_access", "text": "MCP Gateway connected to 45 systems; a network proxy decides whether privileged requests proceed and injects credentials after approval", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-2", "relation": "supports" } ] }, { "id": "sierra-pinecone--architecture-knowledge", "approach_id": "sierra-pinecone", "field": "architecture.knowledge", "text": "Durable sessions over disposable environments; Redis Streams", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" }, { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--architecture-credentials", "approach_id": "sierra-pinecone", "field": "architecture.credentials", "text": "Gateway enforces the invoking employee's access at tool-call time and isolates customer data; the harness never holds the real secret", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-2", "relation": "supports" }, { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--architecture-context-mgmt", "approach_id": "sierra-pinecone", "field": "architecture.context_mgmt", "text": "Durable session state separated from ephemeral compute", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--primitives-0", "approach_id": "sierra-pinecone", "field": "primitives.0", "text": "One gateway spanning 45 systems, enforcing employee permissions at call time", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-2", "relation": "supports" } ] }, { "id": "sierra-pinecone--primitives-1", "approach_id": "sierra-pinecone", "field": "primitives.1", "text": "Recoverable K8s runners with a network proxy for credential injection", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--primitives-2", "approach_id": "sierra-pinecone", "field": "primitives.2", "text": "Routes model and environment by request intent, independent of the tool layer", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" } ] }, { "id": "sierra-pinecone--key-metrics-0", "approach_id": "sierra-pinecone", "field": "key_metrics.0", "text": "More than 75,000 sessions created by 600 people in the month preceding the report", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports", "locator": "Preserved content.md, lines 134–139" } ], "reported_by": "Sierra", "metric_scope": "Pinecone users and sessions in the month preceding the report; calendar month unspecified" }, { "id": "sierra-pinecone--key-metrics-1", "approach_id": "sierra-pinecone", "field": "key_metrics.1", "text": "70% of company PRs opened through Pinecone in the month preceding the report", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports", "locator": "Preserved content.md, lines 138" } ], "reported_by": "Sierra", "metric_scope": "Company PRs opened through Pinecone in the month preceding the report", "denominator": "Sierra PRs opened in that month; not merged PRs" }, { "id": "sierra-pinecone--lessons-learned-0", "approach_id": "sierra-pinecone", "field": "lessons_learned.0", "text": "Collapse departmental bots into one agent; cross-functional jobs don't respect org-chart boundaries", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" } ] }, { "id": "sierra-pinecone--lessons-learned-1", "approach_id": "sierra-pinecone", "field": "lessons_learned.1", "text": "Own the routing/context/workflow layer; let models be interchangeable (different models win at planning, coding, prose)", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" }, { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--lessons-learned-2", "approach_id": "sierra-pinecone", "field": "lessons_learned.2", "text": "Enforce permissions at the tool-call layer via a gateway, not via prompt instructions", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-2", "relation": "supports" }, { "source_id": "sierra-pinecone-source-3", "relation": "supports" } ] }, { "id": "sierra-pinecone--lessons-learned-3", "approach_id": "sierra-pinecone", "field": "lessons_learned.3", "text": "Session counts and tool calls are usage, not value; track business outcomes instead", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" } ] }, { "id": "sierra-pinecone--operating-models-0", "approach_id": "sierra-pinecone", "field": "operating_models.0", "text": "Level 3 for employee request → reviewed agent output; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The source documents delegated work through a company-wide agent without establishing outcome-only supervision.", "valid_at": "2026", "evidence": [ { "source_id": "sierra-pinecone-source-1", "relation": "supports" } ] }, { "id": "slack-context-system--summary", "approach_id": "slack-context-system", "field": "summary", "text": "A coordinator/dispatcher multi-agent design with structured context channels for long-running investigations spanning hundreds of steps.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--headline-metric", "approach_id": "slack-context-system", "field": "headline_metric", "text": "Context management for security investigations spanning hundreds of inference requests and megabytes of output", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports", "locator": "Preserved content.md, lines 34–38" }, { "source_id": "slack-context-system-source-2", "relation": "supports", "locator": "Preserved content.md, lines 22" } ], "reported_by": "Slack", "metric_scope": "Complex security investigations requiring tailored multi-agent context; qualitative workload scale" }, { "id": "slack-context-system--architecture-harness", "approach_id": "slack-context-system", "field": "architecture.harness", "text": "Coordinator/dispatcher: a central coordinator dispatches to expert agents and to critic agents", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--architecture-model", "approach_id": "slack-context-system", "field": "architecture.model", "text": "Not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--architecture-tool-access", "approach_id": "slack-context-system", "field": "architecture.tool_access", "text": "Expert agents produce reports; critic agents evaluate them using evidence-inspection tools", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--architecture-context-mgmt", "approach_id": "slack-context-system", "field": "architecture.context_mgmt", "text": "Three channels; Director's Journal (working memory), Critic's Review (credibility-weighted findings), Critic's Timeline (deduped chronological synthesis)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--primitives-0", "approach_id": "slack-context-system", "field": "primitives.0", "text": "Structured working memory: findings, decisions, questions, hypotheses", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--primitives-1", "approach_id": "slack-context-system", "field": "primitives.1", "text": "A truth filter with credibility scores over submitted findings", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--primitives-2", "approach_id": "slack-context-system", "field": "primitives.2", "text": "A chronological, deduped, conflict-resolved synthesis retained across steps", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--key-metrics-0", "approach_id": "slack-context-system", "field": "key_metrics.0", "text": "Handles multi-agent runs spanning hundreds of requests and megabytes of output", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports", "locator": "Preserved content.md, lines 34–38" }, { "source_id": "slack-context-system-source-2", "relation": "supports", "locator": "Preserved content.md, lines 22" } ], "reported_by": "Slack", "metric_scope": "Complex security investigations requiring tailored multi-agent context; qualitative workload scale" }, { "id": "slack-context-system--lessons-learned-0", "approach_id": "slack-context-system", "field": "lessons_learned.0", "text": "Don't pass all information at every step; build structured summaries agents can reliably build on", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--lessons-learned-1", "approach_id": "slack-context-system", "field": "lessons_learned.1", "text": "Separate expert agents (produce) from critic agents (evaluate); corroborated findings are prioritized", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--lessons-learned-2", "approach_id": "slack-context-system", "field": "lessons_learned.2", "text": "Context management becomes its own subsystem once runs get long", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" }, { "source_id": "slack-context-system-source-2", "relation": "supports" } ] }, { "id": "slack-context-system--operating-models-0", "approach_id": "slack-context-system", "field": "operating_models.0", "text": "Unclassified for long-running investigation → synthesized report; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The research pattern documents agent coordination and criticism but not the normal human attention boundary.", "valid_at": "2026", "evidence": [ { "source_id": "slack-context-system-source-1", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--summary", "approach_id": "spotify-honk-xirp", "field": "summary", "text": "Honk is Spotify's background coding agent (high confidence); Xirp is the workspace/context/session layer around it (medium confidence). Honk runs on a Claude-Agent-SDK harness in Kubernetes, uses trusted CI tools, and combines formatting and linting with LLM-based diff evaluation.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-2", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-3", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--headline-metric", "approach_id": "spotify-honk-xirp", "field": "headline_metric", "text": "1,500+ merged pull requests generated by Honk", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports", "locator": "Preserved content.md, lines 70" } ], "reported_by": "Spotify", "metric_scope": "Cumulative merged Honk-generated PRs reported in Part 1; no measurement cutoff in preserved text", "measurement_method": "Company-reported merged pull request count" }, { "id": "spotify-honk-xirp--architecture-sandbox", "approach_id": "spotify-honk-xirp", "field": "architecture.sandbox", "text": "Honk runs in a constrained Kubernetes container (it does not inherit arbitrary engineer credentials); jobs execute inside Fleet Management / Fleetshift", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-2", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--architecture-harness", "approach_id": "spotify-honk-xirp", "field": "architecture.harness", "text": "Honk: Claude Agent SDK + Spotify's own harness + Kubernetes pods; Honk v2 adds shared sessions, projects, and Chirp orchestration", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-2", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--architecture-model", "approach_id": "spotify-honk-xirp", "field": "architecture.model", "text": "Claude via the Agent SDK (Claude-centric); the surrounding platform is multi-model", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-2", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-3", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--architecture-interfaces", "approach_id": "spotify-honk-xirp", "field": "architecture.interfaces", "text": "slack, github, cli", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--architecture-tool-access", "approach_id": "spotify-honk-xirp", "field": "architecture.tool_access", "text": "Limited, deliberate tool surface; trusted CI tools verify changes, while an internal CLI runs formatting and linting through MCP and evaluates diffs with an LLM judge", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-2", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--architecture-knowledge", "approach_id": "spotify-honk-xirp", "field": "architecture.knowledge", "text": "Backstage/catalog ownership and developer-standardization primitives built over years, exposed via MCP/CLI", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-3", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--architecture-credentials", "approach_id": "spotify-honk-xirp", "field": "architecture.credentials", "text": "Runs in a constrained container rather than inheriting engineer credentials", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-2", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--architecture-context-mgmt", "approach_id": "spotify-honk-xirp", "field": "architecture.context_mgmt", "text": "Fleet Management coordinates repo targeting, builds/tests, and PR workflow at scale; Xirp pairs sessions with Portal org context", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-3", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--primitives-0", "approach_id": "spotify-honk-xirp", "field": "primitives.0", "text": "The harness combines build/test tools, formatting and linting, and LLM-based diff evaluation", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--primitives-1", "approach_id": "spotify-honk-xirp", "field": "primitives.1", "text": "Target thousands of repos, run builds/tests, open and merge PRs at fleet scale", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-2", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--primitives-2", "approach_id": "spotify-honk-xirp", "field": "primitives.2", "text": "The session/context surface around the agent rather than an autonomous worker itself", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-3", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--key-metrics-0", "approach_id": "spotify-honk-xirp", "field": "key_metrics.0", "text": "1,500+ merged AI-generated PRs (Honk)", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports", "locator": "Preserved content.md, lines 70" } ], "reported_by": "Spotify", "metric_scope": "Cumulative merged Honk-generated PRs reported in Part 1; no measurement cutoff in preserved text" }, { "id": "spotify-honk-xirp--key-metrics-1", "approach_id": "spotify-honk-xirp", "field": "key_metrics.1", "text": "60-90% time savings on migrations vs manual", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports", "locator": "Preserved content.md, lines 70–77" } ], "reported_by": "Spotify", "metric_scope": "Selected code migrations using Honk compared with writing the changes manually", "denominator": "Manual completion time for the same migration work" }, { "id": "spotify-honk-xirp--key-metrics-2", "approach_id": "spotify-honk-xirp", "field": "key_metrics.2", "text": "Around half of Spotify PRs automated by Fleet Management since mid-2024, including deterministic transformations", "kind": "metric", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2024", "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports", "locator": "Preserved content.md, lines 22–38" } ], "reported_by": "Spotify", "metric_scope": "Fleet Management automated pull requests since mid-2024, including deterministic transforms", "denominator": "All Spotify pull requests" }, { "id": "spotify-honk-xirp--lessons-learned-0", "approach_id": "spotify-honk-xirp", "field": "lessons_learned.0", "text": "Combine deterministic build, formatting, and lint checks with LLM-based diff evaluation", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--lessons-learned-1", "approach_id": "spotify-honk-xirp", "field": "lessons_learned.1", "text": "Constrain the agent's environment and tools rather than handing it engineer credentials", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--lessons-learned-2", "approach_id": "spotify-honk-xirp", "field": "lessons_learned.2", "text": "Invest in org context (Backstage/catalog) before agentic dev accelerates", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" }, { "source_id": "spotify-honk-xirp-source-3", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--lessons-learned-3", "approach_id": "spotify-honk-xirp", "field": "lessons_learned.3", "text": "Autonomy without structure fragments into per-engineer configs; shared org context is the multiplier", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "spotify-honk-xirp-source-3", "relation": "supports" } ] }, { "id": "spotify-honk-xirp--operating-models-0", "approach_id": "spotify-honk-xirp", "field": "operating_models.0", "text": "Level 3 for Honk coding task → verified pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source documents automated verification followed by a pull-request workflow that retains human review.", "valid_at": "2025", "evidence": [ { "source_id": "spotify-honk-xirp-source-1", "relation": "supports" } ] }, { "id": "stripe-minions--summary", "approach_id": "stripe-minions", "field": "summary", "text": "Homegrown one-shot coding agents that read work context, locate the right repo/workspace, implement a change end-to-end, and produce a PR for human review.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports" }, { "source_id": "stripe-minions-source-3", "relation": "contextualizes", "locator": "Preserved content.md, lines 146–158, 190–194" }, { "source_id": "stripe-minions-source-4", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-5", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-6", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-7", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" } ] }, { "id": "stripe-minions--headline-metric", "approach_id": "stripe-minions", "field": "headline_metric", "text": "Over 1,300 completely minion-produced PRs merged per week in Part 2, with human review and no human-written code", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Stripe reports merged PR volume without an independent count or measurement window; Part 2 is explicitly later than Part 1.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-2", "relation": "supports", "locator": "Preserved content.md, lines 12" }, { "source_id": "stripe-minions-source-3", "relation": "contextualizes", "locator": "Preserved content.md, lines 146–158, 190–194" }, { "source_id": "stripe-minions-source-4", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-5", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-6", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-7", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" } ], "reported_by": "Stripe", "metric_scope": "Weekly merged PRs completely produced by Minions; Part 2 report, up from Part 1" }, { "id": "stripe-minions--architecture-sandbox", "approach_id": "stripe-minions", "field": "architecture.sandbox", "text": "Pre-warmed AWS EC2 devboxes in the QA environment, isolated from real user data, production services, and arbitrary network egress", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-2", "relation": "supports", "locator": "Preserved content.md, lines 24–30, 84" } ] }, { "id": "stripe-minions--architecture-harness", "approach_id": "stripe-minions", "field": "architecture.harness", "text": "Fork of Block's goose, orchestrated by code-defined blueprints that interleave agent loops with deterministic lint, git, and CI steps", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-2", "relation": "supports", "locator": "Preserved content.md, lines 38–56" } ] }, { "id": "stripe-minions--architecture-model", "approach_id": "stripe-minions", "field": "architecture.model", "text": "Not specified", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports" } ] }, { "id": "stripe-minions--architecture-interfaces", "approach_id": "stripe-minions", "field": "architecture.interfaces", "text": "slack, github, cli, web, internal-ui", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports", "locator": "Preserved content.md, lines 34–48" } ] }, { "id": "stripe-minions--architecture-tool-access", "approach_id": "stripe-minions", "field": "architecture.tool_access", "text": "Curated subsets of Toolshed MCP tools for internal documentation, tickets, build status, and code intelligence; security controls constrain destructive actions", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-2", "relation": "supports", "locator": "Preserved content.md, lines 72–84" } ] }, { "id": "stripe-minions--architecture-knowledge", "approach_id": "stripe-minions", "field": "architecture.knowledge", "text": "Repository-scoped rule files shared with human-operated coding agents, plus internal context fetched through MCP", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-2", "relation": "supports", "locator": "Preserved content.md, lines 64–76" } ] }, { "id": "stripe-minions--architecture-credentials", "approach_id": "stripe-minions", "field": "architecture.credentials", "text": "Full permissions inside quarantined devboxes; MCP security controls limit destructive actions, and production pull requests require human review", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-2", "relation": "supports", "locator": "Preserved content.md, lines 40–42, 84" } ] }, { "id": "stripe-minions--primitives-0", "approach_id": "stripe-minions", "field": "primitives.0", "text": "From work context to merge-ready PR in one shot, with human approval", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports" }, { "source_id": "stripe-minions-source-2", "relation": "supports" } ] }, { "id": "stripe-minions--key-metrics-0", "approach_id": "stripe-minions", "field": "key_metrics.0", "text": "Over 1,000 completely minion-produced PRs merged per week in Part 1, with human review and no human-written code", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Stripe reports merged PR volume without an independent count or measurement window; no calendar metric date appears in the capture.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports", "locator": "Preserved content.md, lines 14" }, { "source_id": "stripe-minions-source-3", "relation": "contextualizes", "locator": "Preserved content.md, lines 146–158, 190–194" }, { "source_id": "stripe-minions-source-4", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-5", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-6", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" }, { "source_id": "stripe-minions-source-7", "relation": "contextualizes", "locator": "Preserved content.md, lines 10–14" } ], "reported_by": "Stripe", "metric_scope": "Weekly merged PRs completely produced by Minions; earlier Part 1 report" }, { "id": "stripe-minions--lessons-learned-0", "approach_id": "stripe-minions", "field": "lessons_learned.0", "text": "One-shot end-to-end coding agents work at scale when humans retain review/approval", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports" } ] }, { "id": "stripe-minions--lessons-learned-1", "approach_id": "stripe-minions", "field": "lessons_learned.1", "text": "Let the agent locate the right repo/workspace itself rather than pre-scoping it", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports" }, { "source_id": "stripe-minions-source-2", "relation": "supports" } ] }, { "id": "stripe-minions--operating-models-0", "approach_id": "stripe-minions", "field": "operating_models.0", "text": "Level 3 for work context → merge-ready pull request; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source explicitly states that humans review and approve production pull requests.", "valid_at": "2026-02-20", "evidence": [ { "source_id": "stripe-minions-source-1", "relation": "supports" } ] }, { "id": "uber-coding-agent--summary", "approach_id": "uber-coding-agent", "field": "summary", "text": "Uber's internal coding agent, reported by its CTO as producing roughly 1,800 complete code changes per week.", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "uber-coding-agent-source-1", "relation": "supports" } ] }, { "id": "uber-coding-agent--headline-metric", "approach_id": "uber-coding-agent", "field": "headline_metric", "text": "~1,800 complete code changes per week (~8% of changes)", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "uber-coding-agent-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24–26" } ], "reported_by": "Uber", "metric_scope": "Code changes written entirely by the internal coding agent and human reviewed", "denominator": "All Uber code changes for the reported 8% share" }, { "id": "uber-coding-agent--architecture-sandbox", "approach_id": "uber-coding-agent", "field": "architecture.sandbox", "text": "unknown", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The preserved sources do not document an execution sandbox; unknown does not mean absent.", "valid_at": null, "evidence": [ { "source_id": "uber-coding-agent-source-1", "relation": "supports" } ] }, { "id": "uber-coding-agent--architecture-harness", "approach_id": "uber-coding-agent", "field": "architecture.harness", "text": "Not specified publicly", "kind": "fact", "provenance": "reported", "confidence": "medium", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "uber-coding-agent-source-1", "relation": "supports" } ] }, { "id": "uber-coding-agent--key-metrics-0", "approach_id": "uber-coding-agent", "field": "key_metrics.0", "text": "~1,800 complete code changes per week (~8% of changes at the time)", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "uber-coding-agent-source-1", "relation": "supports", "locator": "Preserved content.md, lines 24–26" } ], "reported_by": "Uber", "metric_scope": "Code changes written entirely by the internal coding agent and human reviewed", "denominator": "All Uber code changes for the reported 8% share" }, { "id": "uber-coding-agent--key-metrics-1", "approach_id": "uber-coding-agent", "field": "key_metrics.1", "text": "95% of engineers use AI tools monthly", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "uber-coding-agent-source-1", "relation": "supports", "locator": "Preserved content.md, lines 18" } ], "reported_by": "Uber", "metric_scope": "Monthly use of AI tools among Uber engineers; not internal-agent-specific adoption", "denominator": "Uber engineers" }, { "id": "uber-coding-agent--operating-models-0", "approach_id": "uber-coding-agent", "field": "operating_models.0", "text": "Unclassified for coding request → complete code change; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The available source reports output volume but does not document where human attention returns.", "valid_at": "2026", "evidence": [ { "source_id": "uber-coding-agent-source-1", "relation": "supports" } ] }, { "id": "uber-ureview--summary", "approach_id": "uber-ureview", "field": "summary", "text": "An event-driven AI code reviewer for Uber's internal review platform that generates, grades, filters, deduplicates, and posts findings while leaving engineers in control of the reviewed change.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Introduction; How It Works" } ] }, { "id": "uber-ureview--headline-metric", "approach_id": "uber-ureview", "field": "headline_metric", "text": "Uber's introduction reports reviews of over 90% of approximately 65,000 weekly diffs, with over 75% usefulness and over 65% addressed comments; a later paragraph says 65,000 diffs per month", "kind": "metric", "provenance": "reported", "confidence": "low", "confidence_reason": "The opening paragraph reports approximately 65,000 weekly diffs, but the cost discussion says 65,000 per month. These conflicting periods remain unresolved; neither is independently verified.", "valid_at": "2025-08", "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Preserved content.md, lines 34, 98" }, { "source_id": "uber-ureview-source-1", "relation": "contradicts", "locator": "Preserved content.md, lines 120 (65,000 per month, versus weekly in line 34)" } ], "reported_by": "Uber", "metric_scope": "Introduction's reported weekly diff coverage, plus comment usefulness/addressed rates; later paragraph's monthly period conflicts and remains unresolved", "denominator": "Introduction reports approximately 65,000 weekly diffs; cost paragraph says monthly. Usefulness covers rated comments; addressed rate covers posted comments", "measurement_method": "Production coverage, developer ratings, and automatic addressed-comment detection" }, { "id": "uber-ureview--architecture-harness", "approach_id": "uber-ureview", "field": "architecture.harness", "text": "A prompt-chained pipeline separates comment generation, confidence grading, validation, semantic deduplication, and category filtering; three specialized assistants were in operation when published", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "How It Works; Comment Generation; Post-Processing" } ] }, { "id": "uber-ureview--architecture-model", "approach_id": "uber-ureview", "field": "architecture.model", "text": "Periodic benchmark evaluation; Claude 4 Sonnet as generator with o4-mini-high as grader was the highest-F1 reported pairing", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Empirical Model Evaluation" } ] }, { "id": "uber-ureview--architecture-interfaces", "approach_id": "uber-ureview", "field": "architecture.interfaces", "text": "internal-ui, ci", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Comment Delivery and Feedback Collection; Impact and Evaluation" } ] }, { "id": "uber-ureview--architecture-tool-access", "approach_id": "uber-ureview", "field": "architecture.tool_access", "text": "Reviews eligible code in Uber's six monorepos across Go, Java, Android, iOS, TypeScript, and Python; richer internal artifacts were not yet connected when published", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Impact and Evaluation; Better at Catching Bugs than Assessing System Design" } ] }, { "id": "uber-ureview--architecture-knowledge", "approach_id": "uber-ureview", "field": "architecture.knowledge", "text": "Surrounding source context plus a shared registry of Uber-specific coding and style rules", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Ingestion and Preprocessing; Comment Generation by Specialized Assistants" } ] }, { "id": "uber-ureview--architecture-context-mgmt", "approach_id": "uber-ureview", "field": "architecture.context_mgmt", "text": "Comments and metadata are streamed through Kafka to Hive for feedback analysis, experiments, and operational dashboards", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Comment Delivery and Feedback Collection" } ] }, { "id": "uber-ureview--primitives-0", "approach_id": "uber-ureview", "field": "primitives.0", "text": "Standard, best-practices, and AppSec reviewers generate findings for different issue classes", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Comment Generation by Specialized Assistants" } ] }, { "id": "uber-ureview--primitives-1", "approach_id": "uber-ureview", "field": "primitives.1", "text": "Confidence grading, semantic deduplication, and historically low-value category suppression reduce noise", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Post-Processing and Quality Filtering" } ] }, { "id": "uber-ureview--primitives-2", "approach_id": "uber-ureview", "field": "primitives.2", "text": "Developer ratings, addressed-comment detection, and a curated benchmark tune prompts, thresholds, and models", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Evaluation and Continuous Improvement" } ] }, { "id": "uber-ureview--key-metrics-0", "approach_id": "uber-ureview", "field": "key_metrics.0", "text": "Introduction reports reviews of over 90% of approximately 65,000 weekly diffs; cost discussion instead says 65,000 diffs per month", "kind": "metric", "provenance": "reported", "confidence": "low", "confidence_reason": "The opening paragraph reports approximately 65,000 weekly diffs, but the cost discussion says 65,000 per month. These conflicting periods remain unresolved; neither is independently verified.", "valid_at": "2025-08", "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Preserved content.md, lines 34" }, { "source_id": "uber-ureview-source-1", "relation": "contradicts", "locator": "Preserved content.md, lines 120 (65,000 per month, versus weekly in line 34)" } ], "reported_by": "Uber", "metric_scope": "Introduction's reported weekly diffs analyzed; monthly period in the cost paragraph conflicts and remains unresolved", "denominator": "Approximately 65,000 weekly diffs according to the introduction; same volume described as monthly in the cost paragraph" }, { "id": "uber-ureview--key-metrics-1", "approach_id": "uber-ureview", "field": "key_metrics.1", "text": "Over 75% of comments rated useful by engineers who interact with the tool", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2025-08", "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Preserved content.md, lines 76, 98" } ], "reported_by": "Uber", "metric_scope": "Comments rated useful by engineers who provide feedback", "denominator": "Comments with engineer interaction", "measurement_method": "Useful / Not Useful rating links" }, { "id": "uber-ureview--key-metrics-2", "approach_id": "uber-ureview", "field": "key_metrics.2", "text": "Over 65% of posted comments addressed in the same changeset", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2025-08", "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Preserved content.md, lines 82, 98" } ], "reported_by": "Uber", "metric_scope": "Posted comments considered addressed in the same changeset", "denominator": "Posted comments", "measurement_method": "Five reruns on the final commit and semantic-similarity matching" }, { "id": "uber-ureview--key-metrics-3", "approach_id": "uber-ureview", "field": "key_metrics.3", "text": "Median review latency of 4 minutes across all six Uber monorepos", "kind": "metric", "provenance": "reported", "confidence": "medium", "confidence_reason": "A linked first-party source states the claim.", "valid_at": "2025-08", "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Preserved content.md, lines 94" } ], "reported_by": "Uber", "metric_scope": "Reviews across all six Uber monorepos", "measurement_method": "Production latency telemetry" }, { "id": "uber-ureview--key-metrics-4", "approach_id": "uber-ureview", "field": "key_metrics.4", "text": "Approximately 1,500 developer hours reportedly saved per week, based on an assumed 10-minute second review per processed commit", "kind": "metric", "provenance": "reported", "confidence": "low", "confidence_reason": "The source estimates about 1,500 hours using over 10,000 commits and a 10-minute assumption; the rounded figures do not arithmetically reconcile exactly and are not observed time savings.", "valid_at": "2025-08", "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Preserved content.md, lines 104" } ], "reported_by": "Uber", "metric_scope": "Processed commits excluding configuration files; modeled second-review time savings", "denominator": "Over 10,000 commits per week", "measurement_method": "Processed commits multiplied by an assumed 10 minutes for a second human review" }, { "id": "uber-ureview--lessons-learned-0", "approach_id": "uber-ureview", "field": "lessons_learned.0", "text": "Prefer fewer high-confidence findings over high comment volume", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The article states this lesson directly and documents the associated mechanisms.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Precision Is More Valuable than Volume" } ] }, { "id": "uber-ureview--lessons-learned-1", "approach_id": "uber-ureview", "field": "lessons_learned.1", "text": "Combine prompts with deterministic filtering, deduplication, evaluation, and feedback instrumentation", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The article states this lesson directly and documents the pipeline.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Guardrails Are Just as Important as Prompts" } ] }, { "id": "uber-ureview--lessons-learned-2", "approach_id": "uber-ureview", "field": "lessons_learned.2", "text": "Roll out gradually by team and assistant while tracking precision, recall, usefulness, and false positives", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The article states this lesson directly and describes the rollout telemetry.", "valid_at": null, "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports", "locator": "Trust Grows with Gradual Rollout" } ] }, { "id": "uber-ureview--operating-models-0", "approach_id": "uber-ureview", "field": "operating_models.0", "text": "Level 3 for pull request → filtered AI review findings; human attention boundary: work-product-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source describes AI-generated review findings while engineers remain in control of the change.", "valid_at": "2025-08-12", "evidence": [ { "source_id": "uber-ureview-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--summary", "approach_id": "workos-project-horizon", "field": "summary", "text": "An internal autonomous 'code factory' where a continuously running swarm of agents handles the implementation loop while engineers focus on requirements and acceptance testing. Deliberately modular so the harness can evolve.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" }, { "source_id": "workos-project-horizon-source-2", "relation": "supports" }, { "source_id": "workos-project-horizon-source-3", "relation": "contextualizes" } ] }, { "id": "workos-project-horizon--architecture-sandbox", "approach_id": "workos-project-horizon", "field": "architecture.sandbox", "text": "Cloudflare Containers + Sandbox SDK; disposable, tightly scoped sandboxes with explicit lifecycle APIs and egress controls; full monorepo stack in Docker dev containers", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--architecture-harness", "approach_id": "workos-project-horizon", "field": "architecture.harness", "text": "Modular by design; the core article runs OpenCode in the sandbox; the Applied AI Showcase runs Claude Remote Routines. The harness is swappable as agent tech changes; separate PM, implementation, and prospective verification/security roles", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" }, { "source_id": "workos-project-horizon-source-2", "relation": "supports" } ] }, { "id": "workos-project-horizon--architecture-model", "approach_id": "workos-project-horizon", "field": "architecture.model", "text": "Swappable; the harness is the constant, not the model", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" }, { "source_id": "workos-project-horizon-source-2", "relation": "supports" } ] }, { "id": "workos-project-horizon--architecture-interfaces", "approach_id": "workos-project-horizon", "field": "architecture.interfaces", "text": "linear, github, slack, web", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--architecture-tool-access", "approach_id": "workos-project-horizon", "field": "architecture.tool_access", "text": "A custom MCP server stitches internal data sources (Datadog, Sentry, Slack, WorkOS Pipes); all outbound traffic proxied through Workers with allowlists, limits, logging, and token injection", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" }, { "source_id": "workos-project-horizon-source-2", "relation": "supports" } ] }, { "id": "workos-project-horizon--architecture-knowledge", "approach_id": "workos-project-horizon", "field": "architecture.knowledge", "text": "AGENTS.md and CLAUDE.md capture scripts, docs, conventions; MCP codifies the patterns engineers already follow; Notion + Figma for specs/mockups", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--architecture-credentials", "approach_id": "workos-project-horizon", "field": "architecture.credentials", "text": "WorkOS Pipes (no OAuth/token-refresh to maintain); scoped short-lived GitHub tokens per user; engineers use their own identity in the MCP; least-privilege + egress controls", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--architecture-context-mgmt", "approach_id": "workos-project-horizon", "field": "architecture.context_mgmt", "text": "The orchestrator pauses/resumes sandboxes and tracks state + artifacts across a run", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--primitives-0", "approach_id": "workos-project-horizon", "field": "primitives.0", "text": "Separate what runs code from what manages the lifecycle", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--primitives-1", "approach_id": "workos-project-horizon", "field": "primitives.1", "text": "A runtime controlled end-to-end, with lifecycle APIs and egress controls for the threat model", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--primitives-2", "approach_id": "workos-project-horizon", "field": "primitives.2", "text": "Each run ships work and produces the next set of fixes, surfacing where the platform is brittle", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--primitives-3", "approach_id": "workos-project-horizon", "field": "primitives.3", "text": "Tuning tools is ongoing, not a one-time integration", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-2", "relation": "supports" } ] }, { "id": "workos-project-horizon--lessons-learned-0", "approach_id": "workos-project-horizon", "field": "lessons_learned.0", "text": "You need purpose-built agent infrastructure; a runtime you control end-to-end with lifecycle APIs and egress controls", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--lessons-learned-1", "approach_id": "workos-project-horizon", "field": "lessons_learned.1", "text": "Separate concerns: sandboxes are an execution primitive; the orchestrator is the control plane", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--lessons-learned-2", "approach_id": "workos-project-horizon", "field": "lessons_learned.2", "text": "Build modularly so the harness can evolve; OpenCode today, Claude Remote Routines tomorrow, without rebuilding the platform", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" }, { "source_id": "workos-project-horizon-source-2", "relation": "supports" } ] }, { "id": "workos-project-horizon--lessons-learned-3", "approach_id": "workos-project-horizon", "field": "lessons_learned.3", "text": "Make autonomy a platform; the system gets faster and more reliable through use as fixes feed back in", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "workos-project-horizon--lessons-learned-4", "approach_id": "workos-project-horizon", "field": "lessons_learned.4", "text": "MCP tuning is an iterative product, not a one-time integration; codify the patterns engineers already follow", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "workos-project-horizon-source-2", "relation": "supports" } ] }, { "id": "workos-project-horizon--operating-models-0", "approach_id": "workos-project-horizon", "field": "operating_models.0", "text": "Level 4 for requirements and acceptance criteria → tested implementation; human attention boundary: outcome-review.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "high", "confidence_reason": "The source describes engineers focusing on requirements and acceptance testing while agents run the implementation loop.", "valid_at": "2026-05-06", "evidence": [ { "source_id": "workos-project-horizon-source-1", "relation": "supports" } ] }, { "id": "ycombinator-agent-infra--summary", "approach_id": "ycombinator-agent-infra", "field": "summary", "text": "Internal agent infrastructure and own harnesses built from the ground up, framed as making AI the operating system the whole organization runs on.", "kind": "fact", "provenance": "reported", "confidence": "low", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "ycombinator-agent-infra-source-1", "relation": "supports" } ] }, { "id": "ycombinator-agent-infra--architecture-sandbox", "approach_id": "ycombinator-agent-infra", "field": "architecture.sandbox", "text": "unknown", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The preserved sources do not document an execution sandbox; unknown does not mean absent.", "valid_at": null, "evidence": [ { "source_id": "ycombinator-agent-infra-source-1", "relation": "supports" } ] }, { "id": "ycombinator-agent-infra--architecture-harness", "approach_id": "ycombinator-agent-infra", "field": "architecture.harness", "text": "Own harnesses built from the ground up for internal AI use", "kind": "fact", "provenance": "reported", "confidence": "low", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "ycombinator-agent-infra-source-1", "relation": "supports" } ] }, { "id": "ycombinator-agent-infra--lessons-learned-0", "approach_id": "ycombinator-agent-infra", "field": "lessons_learned.0", "text": "Don't add AI as a feature; make it the operating system the whole organization runs on", "kind": "inference", "provenance": "catalog-judgment", "confidence": "low", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "ycombinator-agent-infra-source-1", "relation": "supports" } ] }, { "id": "ycombinator-agent-infra--lessons-learned-1", "approach_id": "ycombinator-agent-infra", "field": "lessons_learned.1", "text": "Build the harness and surrounding infra in-house rather than bolting onto a hosted agent", "kind": "inference", "provenance": "catalog-judgment", "confidence": "low", "confidence_reason": "Only limited public evidence supports this claim.", "valid_at": null, "evidence": [ { "source_id": "ycombinator-agent-infra-source-1", "relation": "supports" } ] }, { "id": "ycombinator-agent-infra--operating-models-0", "approach_id": "ycombinator-agent-infra", "field": "operating_models.0", "text": "Unclassified for internal request → agent-assisted organizational work; human attention boundary: unknown.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "unverified", "confidence_reason": "The source describes internal agent infrastructure but not a sufficiently specific human attention boundary.", "valid_at": "2026", "evidence": [ { "source_id": "ycombinator-agent-infra-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--summary", "approach_id": "zup-codegen", "field": "summary", "text": "A research-documented internal coding agent where constrained editing tools and layered safety controls mattered more than prompt tweaks.", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--architecture-harness", "approach_id": "zup-codegen", "field": "architecture.harness", "text": "Constrained editing tools; state-management concerns; progressive levels of human oversight", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--architecture-model", "approach_id": "zup-codegen", "field": "architecture.model", "text": "Not specified (see paper)", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--architecture-tool-access", "approach_id": "zup-codegen", "field": "architecture.tool_access", "text": "Constrained editing tools rather than free-form code generation", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--architecture-credentials", "approach_id": "zup-codegen", "field": "architecture.credentials", "text": "Layered safety controls", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--primitives-0", "approach_id": "zup-codegen", "field": "primitives.0", "text": "Tools that limit what the agent can change, vs. unconstrained generation", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--primitives-1", "approach_id": "zup-codegen", "field": "primitives.1", "text": "Levels of human review that increase trust over time", "kind": "fact", "provenance": "reported", "confidence": "high", "confidence_reason": "A linked first-party source states the claim.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--lessons-learned-0", "approach_id": "zup-codegen", "field": "lessons_learned.0", "text": "Targeted tool design and layered safety controls matter more than prompt tweaks", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--lessons-learned-1", "approach_id": "zup-codegen", "field": "lessons_learned.1", "text": "Progressive levels of human oversight help build trust before granting autonomy", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The catalog derives this observation from the linked sources.", "valid_at": null, "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] }, { "id": "zup-codegen--operating-models-0", "approach_id": "zup-codegen", "field": "operating_models.0", "text": "Level 2 for constrained coding task → human-supervised edit; human attention boundary: continuous-steering.", "kind": "inference", "provenance": "catalog-judgment", "confidence": "medium", "confidence_reason": "The research source describes progressive human oversight but does not establish background delegation.", "valid_at": "2026", "evidence": [ { "source_id": "zup-codegen-source-1", "relation": "supports" } ] } ], "sources": [ { "id": "airbnb-airchat-source-1", "title": "Agentic coding at Airbnb (DPE.org)", "url": "https://dpe.org/sessions/szczepan-faber-mike-nakhimovich/agentic-coding-at-airbnb/", "canonical_url": "https://dpe.org/sessions/szczepan-faber-mike-nakhimovich/agentic-coding-at-airbnb/", "kind": "talk", "provenance_class": "direct-participant", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "airbnb-airchat-source-1", "original_url": "https://dpe.org/sessions/szczepan-faber-mike-nakhimovich/agentic-coding-at-airbnb/", "final_url": "https://dpe.org/sessions/szczepan-faber-mike-nakhimovich/agentic-coding-at-airbnb/", "captured_at": "2026-08-31T17:38:40Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/airbnb-airchat-source-1/content.md", "sha256": "sha256:4ab4a7c2fdbc723d88e0df3f7d7582d1a8c37f143a1627c17c5f602c5941d862", "bytes": 1307 } } }, "approach_id": "airbnb-airchat" }, { "id": "airbnb-airchat-source-2", "title": "Beyond the CLI (DX podcast)", "url": "https://getdx.com/podcast/beyond-the-cli-agentic-ai-for-async-workloads-and-non-developers/", "canonical_url": "https://getdx.com/podcast/beyond-the-cli-agentic-ai-for-async-workloads-and-non-developers/", "kind": "podcast", "provenance_class": "direct-participant", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "airbnb-airchat-source-2", "original_url": "https://getdx.com/podcast/beyond-the-cli-agentic-ai-for-async-workloads-and-non-developers/", "final_url": "https://getdx.com/podcast/beyond-the-cli-agentic-ai-for-async-workloads-and-non-developers/", "captured_at": "2026-08-31T17:38:48Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/airbnb-airchat-source-2/content.md", "sha256": "sha256:f5c2460670c198c7a0822d397e83c916849b91f0309eb79dcf4bb40e11e510a1", "bytes": 50796 } } }, "approach_id": "airbnb-airchat" }, { "id": "airbnb-airchat-source-3", "title": "How to get your team past the AI (The AI Thinker)", "url": "https://www.theaithinker.com/p/how-to-get-your-team-past-the-ai", "canonical_url": "https://www.theaithinker.com/p/how-to-get-your-team-past-the-ai", "kind": "news", "provenance_class": "independent-secondary", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "airbnb-airchat-source-3", "original_url": "https://www.theaithinker.com/p/how-to-get-your-team-past-the-ai", "final_url": "https://www.theaithinker.com/p/how-to-get-your-team-past-the-ai", "captured_at": "2026-08-31T17:38:58Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/airbnb-airchat-source-3/content.md", "sha256": "sha256:1d1dc23408c8f7e01428147e7091b605620e8fe66d637f61fa743b85bdf1c7c6", "bytes": 34229 } } }, "approach_id": "airbnb-airchat" }, { "id": "atlassian-rovo-dev-source-1", "title": "Improving the coding agent experience", "url": "https://www.atlassian.com/blog/atlassian-engineering/improving-coding-agent-experience", "canonical_url": "https://www.atlassian.com/blog/atlassian-engineering/improving-coding-agent-experience", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "atlassian-rovo-dev-source-1", "original_url": "https://www.atlassian.com/blog/atlassian-engineering/improving-coding-agent-experience", "final_url": "https://www.atlassian.com/blog/atlassian-engineering/improving-coding-agent-experience", "captured_at": "2026-08-31T17:39:24Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/atlassian-rovo-dev-source-1/content.md", "sha256": "sha256:a468f0d31e7772c927f2d53ba3086ecd78871bc3ec0a73ac332278804eb6b1da", "bytes": 10923 } } }, "approach_id": "atlassian-rovo-dev" }, { "id": "atlassian-rovo-dev-source-2", "title": "Developer productivity improved with Rovo Dev", "url": "https://www.atlassian.com/blog/atlassian-engineering/developer-productivity-improved-with-rovo-dev", "canonical_url": "https://www.atlassian.com/blog/atlassian-engineering/developer-productivity-improved-with-rovo-dev", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "atlassian-rovo-dev-source-2", "original_url": "https://www.atlassian.com/blog/atlassian-engineering/developer-productivity-improved-with-rovo-dev", "final_url": "https://www.atlassian.com/blog/atlassian-engineering/developer-productivity-improved-with-rovo-dev", "captured_at": "2026-08-31T17:39:33Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/atlassian-rovo-dev-source-2/content.md", "sha256": "sha256:b1647b8ac1a5dbd5c6e8dee67b596d668681746db9a3374165a637b11e1607f7", "bytes": 12256 } } }, "approach_id": "atlassian-rovo-dev" }, { "id": "atlassian-rovo-dev-source-3", "title": "HULA: Human-in-the-loop software development agents (arXiv 2411.12924)", "url": "https://arxiv.org/abs/2411.12924", "canonical_url": "https://arxiv.org/abs/2411.12924", "kind": "paper", "provenance_class": "first-party", "role": "evidence", "published_at": "2024-11", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "atlassian-rovo-dev-source-3", "original_url": "https://arxiv.org/abs/2411.12924", "final_url": "https://arxiv.org/abs/2411.12924", "captured_at": "2026-08-31T17:39:42Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/atlassian-rovo-dev-source-3/content.md", "sha256": "sha256:a5daa4616710f897d08ae2b3798dc485ca57ec5171a1b7bce536bb78472489f7", "bytes": 2543 } } }, "approach_id": "atlassian-rovo-dev" }, { "id": "block-builderbot-source-1", "title": "Block rolls out Builderbot, a new suite of AI-native tools that changes the way we ship", "url": "https://block.xyz/inside/block-rolls-out-builderbot-a-new-suite-of-ai-native-tools-that-changes-the-way-we-ship", "canonical_url": "https://block.xyz/inside/block-rolls-out-builderbot-a-new-suite-of-ai-native-tools-that-changes-the-way-we-ship", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "block-builderbot-source-1", "original_url": "https://block.xyz/inside/block-rolls-out-builderbot-a-new-suite-of-ai-native-tools-that-changes-the-way-we-ship", "final_url": "https://block.xyz/inside/block-rolls-out-builderbot-a-new-suite-of-ai-native-tools-that-changes-the-way-we-ship", "captured_at": "2026-08-31T17:39:51Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/block-builderbot-source-1/content.md", "sha256": "sha256:4189377fdd7349ee5fb0b8aa176460f51543348c6d53b1ae7e8a20943818643f", "bytes": 4286 } } }, "role": "evidence", "approach_id": "block-builderbot" }, { "id": "block-builderbot-source-2", "title": "Protecting our systems with intelligence (Builderbot architecture)", "url": "https://engineering.block.xyz/blog/protecting-our-systems-with-intelligence", "canonical_url": "https://engineering.block.xyz/blog/protecting-our-systems-with-intelligence", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "block-builderbot-source-2", "original_url": "https://engineering.block.xyz/blog/protecting-our-systems-with-intelligence", "final_url": "https://engineering.block.xyz/blog/protecting-our-systems-with-intelligence", "captured_at": "2026-08-31T17:39:58Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/block-builderbot-source-2/content.md", "sha256": "sha256:2e7d3fb1c1682165745f9b78ef2560e0713e182eebd59887768cac12aa7dc9f7", "bytes": 9784 } } }, "role": "evidence", "approach_id": "block-builderbot" }, { "id": "block-builderbot-source-3", "title": "block/builderbot source repository", "url": "https://github.com/block/builderbot", "canonical_url": "https://github.com/block/builderbot", "kind": "repository", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "block-builderbot-source-3", "original_url": "https://github.com/block/builderbot", "final_url": "https://github.com/block/builderbot", "captured_at": "2026-08-31T17:40:09Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/block-builderbot-source-3/content.md", "sha256": "sha256:503f6115efd4f7042b22085d685af3ca026c145e8efe57544be7e8776f0d0de6", "bytes": 2503 } } }, "approach_id": "block-builderbot" }, { "id": "block-builderbot-source-4", "title": "Hacker News discussion of the Builderbot announcement", "url": "https://news.ycombinator.com/item?id=48618973", "canonical_url": "https://news.ycombinator.com/item?id=48618973", "kind": "hn-thread", "provenance_class": "community", "role": "commentary", "authors": [ "msolujic" ], "published_at": "2026-06-21", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "block-builderbot-source-4", "original_url": "https://news.ycombinator.com/item?id=48618973", "final_url": "https://news.ycombinator.com/item?id=48618973", "captured_at": "2026-08-31T17:40:18Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/block-builderbot-source-4/content.md", "sha256": "sha256:23805ab1d4e83b1cfb9d18ba3cdbcdd49851b1276d05ac826a63d013334dac13", "bytes": 836 } } }, "approach_id": "block-builderbot" }, { "id": "block-builderbot-source-5", "title": "Hacker News comment that points to the Builderbot repository", "url": "https://news.ycombinator.com/item?id=48619052", "canonical_url": "https://news.ycombinator.com/item?id=48619052", "kind": "hn-comment", "provenance_class": "community", "role": "discovery", "authors": [ "kha1n3vol3" ], "published_at": "2026-06-21", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "block-builderbot-source-5", "original_url": "https://news.ycombinator.com/item?id=48619052", "final_url": "https://news.ycombinator.com/item?id=48619052", "captured_at": "2026-08-31T17:40:26Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/block-builderbot-source-5/content.md", "sha256": "sha256:1073d4b99c35b7feb054dc79e8a8ad71267aabbed5735d4972f5c26c62f6fe4d", "bytes": 560 } } }, "approach_id": "block-builderbot" }, { "id": "brex-agent-platform-source-1", "title": "Agent, Human, Ops: How Brex Is Changing Roles and Workflows", "url": "https://www.firstround.com/ai/brex", "canonical_url": "https://www.firstround.com/ai/brex", "kind": "case-study", "provenance_class": "independent-secondary", "published_at": "2025-09-25", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "brex-agent-platform-source-1", "original_url": "https://www.firstround.com/ai/brex", "final_url": "https://www.firstround.com/ai/brex", "captured_at": "2026-08-31T17:40:35Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/brex-agent-platform-source-1/content.md", "sha256": "sha256:7f93ca0f790cfee533cc68de90faaafb87555858453c686eb9a67e408c56d428", "bytes": 26716 } } }, "role": "evidence", "approach_id": "brex-agent-platform" }, { "id": "browserbase-bb-source-1", "title": "How we build internal agents at Browserbase", "url": "https://browserbase.com/blog/internal-agents", "canonical_url": "https://browserbase.com/blog/internal-agents", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "browserbase-bb-source-1", "original_url": "https://browserbase.com/blog/internal-agents", "final_url": "https://www.browserbase.com/blog/internal-agents", "captured_at": "2026-08-31T17:40:42Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/browserbase-bb-source-1/content.md", "sha256": "sha256:dd3ad8521c55701e44595b24ddfdda6f15ed6a3d9c47b7fef4893ae8b92858e9", "bytes": 21167 } } }, "role": "evidence", "approach_id": "browserbase-bb" }, { "id": "cloudflare-ai-stack-source-1", "title": "The AI engineering stack we built internally", "url": "https://blog.cloudflare.com/internal-ai-engineering-stack/", "canonical_url": "https://blog.cloudflare.com/internal-ai-engineering-stack/", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "cloudflare-ai-stack-source-1", "original_url": "https://blog.cloudflare.com/internal-ai-engineering-stack/", "final_url": "https://blog.cloudflare.com/internal-ai-engineering-stack/", "captured_at": "2026-08-31T17:40:51Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/cloudflare-ai-stack-source-1/content.md", "sha256": "sha256:2454a2ab695ab9611aa982b541b698f3bb948358a6680c7da13e66ad7307a9d8", "bytes": 29257 } } }, "role": "evidence", "approach_id": "cloudflare-ai-stack" }, { "id": "cloudflare-ai-stack-source-2", "title": "Hacker News discussion of Cloudflare's internal AI engineering stack", "url": "https://news.ycombinator.com/item?id=47837240", "canonical_url": "https://news.ycombinator.com/item?id=47837240", "kind": "hn-thread", "provenance_class": "community", "role": "commentary", "authors": [ "mavelikara" ], "published_at": "2026-04-20", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "cloudflare-ai-stack-source-2", "original_url": "https://news.ycombinator.com/item?id=47837240", "final_url": "https://news.ycombinator.com/item?id=47837240", "captured_at": "2026-08-31T17:40:59Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/cloudflare-ai-stack-source-2/content.md", "sha256": "sha256:529a706a81204c6f55ca7acf5ff98d9343c7060b986a6dcf535f579dfe4ef0dc", "bytes": 742 } } }, "approach_id": "cloudflare-ai-stack" }, { "id": "coinbase-forge-mux-source-1", "title": "Coding had a concurrency problem: how Mux helped solve it", "url": "https://www.coinbase.com/de/blog/coding-had-a-concurrency-problem-how-mux-helped-solve-it", "canonical_url": "https://www.coinbase.com/de/blog/coding-had-a-concurrency-problem-how-mux-helped-solve-it", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "coinbase-forge-mux-source-1", "original_url": "https://www.coinbase.com/de/blog/coding-had-a-concurrency-problem-how-mux-helped-solve-it", "final_url": "https://www.coinbase.com/de/blog/coding-had-a-concurrency-problem-how-mux-helped-solve-it", "captured_at": "2026-08-31T17:41:07Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/coinbase-forge-mux-source-1/content.md", "sha256": "sha256:741948a748f28315d02d2fe5e5d6e2816d755179592e2cbf5c6b57be6daf8850", "bytes": 6767 } } }, "role": "evidence", "approach_id": "coinbase-forge-mux" }, { "id": "coinbase-forge-mux-source-2", "title": "Coinbase × Linear (Forge workflow)", "url": "https://linear.app/customers/coinbase", "canonical_url": "https://linear.app/customers/coinbase", "kind": "case-study", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "coinbase-forge-mux-source-2", "original_url": "https://linear.app/customers/coinbase", "final_url": "https://linear.app/customers/coinbase", "captured_at": "2026-08-31T17:41:18Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/coinbase-forge-mux-source-2/content.md", "sha256": "sha256:b1b946c51a9f2ba6967b8d9a0ebd11f303714242714f4c22f8c09909f0f97f89", "bytes": 4437 } } }, "role": "evidence", "approach_id": "coinbase-forge-mux" }, { "id": "databricks-costar-source-1", "title": "coSTAR: how we ship AI agents at Databricks fast", "url": "https://www.databricks.com/blog/costar-how-we-ship-ai-agents-databricks-fast-without-breaking-things", "canonical_url": "https://www.databricks.com/blog/costar-how-we-ship-ai-agents-databricks-fast-without-breaking-things", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "databricks-costar-source-1", "original_url": "https://www.databricks.com/blog/costar-how-we-ship-ai-agents-databricks-fast-without-breaking-things", "final_url": "https://www.databricks.com/blog/costar-how-we-ship-ai-agents-databricks-fast-without-breaking-things", "captured_at": "2026-08-31T17:41:27Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/databricks-costar-source-1/content.md", "sha256": "sha256:01d6e070fbb6459b6186e421c6c0c61e49e83cf316dbe42d9a801266114bce1f", "bytes": 24383 } } }, "approach_id": "databricks-costar" }, { "id": "databricks-costar-source-2", "title": "Benchmarking coding agents on a multi-million line codebase", "url": "https://www.databricks.com/blog/benchmarking-coding-agents-databricks-multi-million-line-codebase", "canonical_url": "https://www.databricks.com/blog/benchmarking-coding-agents-databricks-multi-million-line-codebase", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "databricks-costar-source-2", "original_url": "https://www.databricks.com/blog/benchmarking-coding-agents-databricks-multi-million-line-codebase", "final_url": "https://www.databricks.com/blog/benchmarking-coding-agents-databricks-multi-million-line-codebase", "captured_at": "2026-08-31T17:41:38Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/databricks-costar-source-2/content.md", "sha256": "sha256:5b765f7c350ef567ed0c8444f52167f0171cc23bd3c48d5d7b632663b3859e78", "bytes": 13678 } } }, "approach_id": "databricks-costar" }, { "id": "domu-clementino-source-1", "title": "Why we split our internal agent in two", "url": "https://domu.ai/blog/why-we-split-our-internal-agent-in-two", "canonical_url": "https://domu.ai/blog/why-we-split-our-internal-agent-in-two", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "domu-clementino-source-1", "original_url": "https://domu.ai/blog/why-we-split-our-internal-agent-in-two", "final_url": "https://domu.ai/blog/why-we-split-our-internal-agent-in-two", "captured_at": "2026-08-31T17:41:45Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/domu-clementino-source-1/content.md", "sha256": "sha256:911788d9acd6aa775fee843f162a7f28ae679631cf1c565a4b790a5bb913ebae", "bytes": 8174 } } }, "role": "evidence", "approach_id": "domu-clementino" }, { "id": "doordash-code-review-source-1", "title": "How DoorDash built an AI code reviewer engineers actually listen to", "url": "https://careersatdoordash.com/blog/doordash-built-an-ai-code-reviewer-engineers-actually-listen-to/", "canonical_url": "https://careersatdoordash.com/blog/doordash-built-an-ai-code-reviewer-engineers-actually-listen-to/", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "doordash-code-review-source-1", "original_url": "https://careersatdoordash.com/blog/doordash-built-an-ai-code-reviewer-engineers-actually-listen-to/", "final_url": "https://careersatdoordash.com/blog/doordash-built-an-ai-code-reviewer-engineers-actually-listen-to/", "captured_at": "2026-08-31T17:41:52Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/doordash-code-review-source-1/content.md", "sha256": "sha256:0dfaa3b3595d97ec0400b6c7570c62e31288a228121cdb618ed30c84585971f4", "bytes": 19641 } } }, "role": "evidence", "approach_id": "doordash-code-review" }, { "id": "doordash-flux-source-1", "title": "Delegating Engineering Work To Cloud-Based Agents (Flux)", "url": "https://x.com/AIatDoorDash/status/2087285008906240193", "canonical_url": "https://x.com/AIatDoorDash/status/2087285008906240193", "kind": "social-post", "provenance_class": "direct-participant", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "doordash-flux-source-1", "original_url": "https://x.com/AIatDoorDash/status/2087285008906240193", "final_url": "https://x.com/AIatDoorDash/status/2087285008906240193", "captured_at": "2026-08-31T17:42:01Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/doordash-flux-source-1/content.md", "sha256": "sha256:3fe095b0e92562a9dbf25d6fc3d6d816ed51168d964ff238d7acf2e455bfb01d", "bytes": 1178 } } }, "role": "evidence", "approach_id": "doordash-flux" }, { "id": "doordash-flux-source-2", "title": "Beyond single agents: DoorDash's collaborative AI ecosystem", "url": "https://careersatdoordash.com/blog/beyond-single-agents-doordash-building-collaborative-ai-ecosystem/", "canonical_url": "https://careersatdoordash.com/blog/beyond-single-agents-doordash-building-collaborative-ai-ecosystem/", "kind": "engineering-blog", "provenance_class": "first-party", "published_at": "2025-11-11", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "doordash-flux-source-2", "original_url": "https://careersatdoordash.com/blog/beyond-single-agents-doordash-building-collaborative-ai-ecosystem/", "final_url": "https://careersatdoordash.com/blog/beyond-single-agents-doordash-building-collaborative-ai-ecosystem/", "captured_at": "2026-08-31T17:55:30Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/doordash-flux-source-2/content.md", "sha256": "sha256:8e9f42ae8559f6dafaa322262329fd03fc96f6cca949ec64a1497cb971d34d34", "bytes": 22418 } } }, "role": "evidence", "approach_id": "doordash-flux" }, { "id": "doordash-flux-source-3", "title": "Delegating Engineering Work To Cloud-Based Agents", "url": "https://careersatdoordash.com/blog/delegating-engineering-work-to-cloud-based-agents/", "canonical_url": "https://careersatdoordash.com/blog/delegating-engineering-work-to-cloud-based-agents/", "kind": "engineering-blog", "provenance_class": "first-party", "published_at": "2026-08-11", "accessed_at": "2026-08-31", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "doordash-flux-source-3", "original_url": "https://careersatdoordash.com/blog/delegating-engineering-work-to-cloud-based-agents/", "final_url": "https://careersatdoordash.com/blog/delegating-engineering-work-to-cloud-based-agents/", "captured_at": "2026-08-31T18:09:17Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/doordash-flux-source-3/content.md", "sha256": "sha256:b319a0ace131d82a2c4f8b477df5511412e3a1c3127e5d74cdb5598624f75b15", "bytes": 9896 } } }, "role": "evidence", "approach_id": "doordash-flux" }, { "id": "dropbox-nova-source-1", "title": "Introducing Nova, our internal platform for coding agents", "url": "https://dropbox.tech/machine-learning/introducing-nova-our-internal-platform-for-coding-agents", "canonical_url": "https://dropbox.tech/machine-learning/introducing-nova-our-internal-platform-for-coding-agents", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "dropbox-nova-source-1", "original_url": "https://dropbox.tech/machine-learning/introducing-nova-our-internal-platform-for-coding-agents", "final_url": "https://dropbox.tech/machine-learning/introducing-nova-our-internal-platform-for-coding-agents", "captured_at": "2026-08-31T17:42:18Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/dropbox-nova-source-1/content.md", "sha256": "sha256:e6b443ee09b9dbe68c0c10aab951ba8f75640d0202f0be4e95fca23f9d3d3593", "bytes": 13723 } } }, "role": "evidence", "approach_id": "dropbox-nova" }, { "id": "dropbox-nova-source-2", "title": "Hacker News submission for Nova", "url": "https://news.ycombinator.com/item?id=48235065", "canonical_url": "https://news.ycombinator.com/item?id=48235065", "kind": "hn-thread", "provenance_class": "community", "role": "discovery", "authors": [ "el_duderino" ], "published_at": "2026-05-22", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "dropbox-nova-source-2", "original_url": "https://news.ycombinator.com/item?id=48235065", "final_url": "https://news.ycombinator.com/item?id=48235065", "captured_at": "2026-08-31T17:42:27Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/dropbox-nova-source-2/content.md", "sha256": "sha256:b608472ab9489885d07bfaa57186859b3a7d4c54b1168ca7a2ab789f5d0f2f8f", "bytes": 886 } } }, "approach_id": "dropbox-nova" }, { "id": "flex-investigation-agent-source-1", "title": "The Flex AI Investigation Agent for HSA/FSA payments", "url": "https://www.withflex.com/blog/the-flex-ai-investigation-agent-for-hsa-fsa-payments", "canonical_url": "https://www.withflex.com/blog/the-flex-ai-investigation-agent-for-hsa-fsa-payments", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "flex-investigation-agent-source-1", "original_url": "https://www.withflex.com/blog/the-flex-ai-investigation-agent-for-hsa-fsa-payments", "final_url": "https://www.withflex.com/blog/the-flex-ai-investigation-agent-for-hsa-fsa-payments", "captured_at": "2026-08-31T17:42:35Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/flex-investigation-agent-source-1/content.md", "sha256": "sha256:cb676fc287fdc54f924f201b15c39e51cef37e925f76c551c471bddc5d035271", "bytes": 14791 } } }, "role": "evidence", "approach_id": "flex-investigation-agent" }, { "id": "github-qubot-source-1", "title": "How we built an internal data analytics agent", "url": "https://github.blog/ai-and-ml/github-copilot/how-we-built-an-internal-data-analytics-agent/", "canonical_url": "https://github.blog/ai-and-ml/github-copilot/how-we-built-an-internal-data-analytics-agent/", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "published_at": "2026-06-19", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "github-qubot-source-1", "original_url": "https://github.blog/ai-and-ml/github-copilot/how-we-built-an-internal-data-analytics-agent/", "final_url": "https://github.blog/ai-and-ml/github-copilot/how-we-built-an-internal-data-analytics-agent/", "captured_at": "2026-08-31T17:42:43Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/github-qubot-source-1/content.md", "sha256": "sha256:820488e2b11a401e35bac3141b644a3f94d5a09e4e3b074c6c5e40d5421992da", "bytes": 8226 } } }, "approach_id": "github-qubot" }, { "id": "harvey-spectre-source-1", "title": "Building Spectre; internal collaborative cloud agent platform", "url": "https://www.harvey.ai/blog/building-spectre-internal-collaborative-cloud-agent-platform", "canonical_url": "https://www.harvey.ai/blog/building-spectre-internal-collaborative-cloud-agent-platform", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "harvey-spectre-source-1", "original_url": "https://www.harvey.ai/blog/building-spectre-internal-collaborative-cloud-agent-platform", "final_url": "https://www.harvey.ai/blog/building-spectre-internal-collaborative-cloud-agent-platform", "captured_at": "2026-08-31T17:42:52Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/harvey-spectre-source-1/content.md", "sha256": "sha256:63f96d247d6786af98fc3af6ffd0bdd7d6826986e70c3a0df9f0bb76adca258d", "bytes": 17423 } } }, "role": "evidence", "approach_id": "harvey-spectre" }, { "id": "harvey-spectre-source-2", "title": "Building an agentic security operations center", "url": "https://www.harvey.ai/blog/building-an-agentic-security-operations-center", "canonical_url": "https://www.harvey.ai/blog/building-an-agentic-security-operations-center", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "harvey-spectre-source-2", "original_url": "https://www.harvey.ai/blog/building-an-agentic-security-operations-center", "final_url": "https://www.harvey.ai/blog/building-an-agentic-security-operations-center", "captured_at": "2026-08-31T17:43:00Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/harvey-spectre-source-2/content.md", "sha256": "sha256:3b8d8ad6234571e9b71c813dce98dc5dd1b990054a16a9aa4f57e7e507b93daf", "bytes": 17193 } } }, "role": "evidence", "approach_id": "harvey-spectre" }, { "id": "hubspot-sidekick-source-1", "title": "Automated code review, the 6-month evolution", "url": "https://product.hubspot.com/blog/automated-code-review-the-6-month-evolution", "canonical_url": "https://product.hubspot.com/blog/automated-code-review-the-6-month-evolution", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "published_at": "2026-03", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "hubspot-sidekick-source-1", "original_url": "https://product.hubspot.com/blog/automated-code-review-the-6-month-evolution", "final_url": "https://product.hubspot.com/blog/automated-code-review-the-6-month-evolution", "captured_at": "2026-08-31T17:43:10Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/hubspot-sidekick-source-1/content.md", "sha256": "sha256:95527b1030126f0b7b832b492bf66b59c3d1297025443f5b1c11841f42ff6326", "bytes": 12018 } } }, "approach_id": "hubspot-sidekick" }, { "id": "hubspot-sidekick-source-2", "title": "Cloud coding agents at HubSpot", "url": "https://product.hubspot.com/blog/cloud-coding-agents-at-hubspot", "canonical_url": "https://product.hubspot.com/blog/cloud-coding-agents-at-hubspot", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "hubspot-sidekick-source-2", "original_url": "https://product.hubspot.com/blog/cloud-coding-agents-at-hubspot", "final_url": "https://product.hubspot.com/blog/cloud-coding-agents-at-hubspot", "captured_at": "2026-08-31T17:43:18Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/hubspot-sidekick-source-2/content.md", "sha256": "sha256:33f147cb1f3574a133c48933b3f34dd167c27a61164afcf7a1d1c2bec1ed4241", "bytes": 14509 } } }, "approach_id": "hubspot-sidekick" }, { "id": "linear-agent-source-1", "title": "How we built Linear Agent", "url": "https://linear.app/now/how-we-built-linear-agent", "canonical_url": "https://linear.app/now/how-we-built-linear-agent", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "linear-agent-source-1", "original_url": "https://linear.app/now/how-we-built-linear-agent", "final_url": "https://linear.app/now/how-we-built-linear-agent", "captured_at": "2026-08-31T17:43:26Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/linear-agent-source-1/content.md", "sha256": "sha256:2742071677e6b38b0bc95644570642d7ed1c92b9b433225e2353b62d29068f86", "bytes": 11316 } } }, "role": "evidence", "approach_id": "linear-agent" }, { "id": "linear-agent-source-2", "title": "How we use Linear Agent at Linear", "url": "https://linear.app/now/how-we-use-linear-agent-at-linear", "canonical_url": "https://linear.app/now/how-we-use-linear-agent-at-linear", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "linear-agent-source-2", "original_url": "https://linear.app/now/how-we-use-linear-agent-at-linear", "final_url": "https://linear.app/now/how-we-use-linear-agent-at-linear", "captured_at": "2026-08-31T17:43:34Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/linear-agent-source-2/content.md", "sha256": "sha256:174026b81583bc2125db5be844623c7b401bfea4f22aa6cd610c3096f3318e60", "bytes": 13512 } } }, "role": "evidence", "approach_id": "linear-agent" }, { "id": "linear-agent-source-3", "title": "Our approach to building the Agent Interaction SDK", "url": "https://linear.app/now/our-approach-to-building-the-agent-interaction-sdk", "canonical_url": "https://linear.app/now/our-approach-to-building-the-agent-interaction-sdk", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "linear-agent-source-3", "original_url": "https://linear.app/now/our-approach-to-building-the-agent-interaction-sdk", "final_url": "https://linear.app/now/our-approach-to-building-the-agent-interaction-sdk", "captured_at": "2026-08-31T17:43:42Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/linear-agent-source-3/content.md", "sha256": "sha256:5ac5435ddad694ca34bcfd20c99d5e21b291f56d21baaaaf538d93bb1f46ef45", "bytes": 11080 } } }, "role": "evidence", "approach_id": "linear-agent" }, { "id": "linear-agent-source-4", "title": "Hacker News submission for how Linear built its agent", "url": "https://news.ycombinator.com/item?id=49252304", "canonical_url": "https://news.ycombinator.com/item?id=49252304", "kind": "hn-thread", "provenance_class": "community", "role": "discovery", "authors": [ "cristinacordova" ], "published_at": "2026-08-11", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "linear-agent-source-4", "original_url": "https://news.ycombinator.com/item?id=49252304", "final_url": "https://news.ycombinator.com/item?id=49252304", "captured_at": "2026-08-31T17:43:49Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/linear-agent-source-4/content.md", "sha256": "sha256:21c31402044279083ba9754cf35e8dd51bd4ba6d2b035970b622555369a81903", "bytes": 393 } } }, "approach_id": "linear-agent" }, { "id": "linear-agent-source-5", "title": "Hacker News submission for the Linear Agent public beta", "url": "https://news.ycombinator.com/item?id=48503334", "canonical_url": "https://news.ycombinator.com/item?id=48503334", "kind": "hn-thread", "provenance_class": "community", "role": "discovery", "authors": [ "samber" ], "published_at": "2026-06-12", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "linear-agent-source-5", "original_url": "https://news.ycombinator.com/item?id=48503334", "final_url": "https://news.ycombinator.com/item?id=48503334", "captured_at": "2026-08-31T17:43:58Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/linear-agent-source-5/content.md", "sha256": "sha256:dde2e308c39473792c911b768fc93dd64f5d3ffefb5f36cef44907ecdfecd4ca", "bytes": 412 } } }, "approach_id": "linear-agent" }, { "id": "linear-agent-source-6", "title": "Introducing Linear Agent", "url": "https://linear.app/changelog/2026-03-24-introducing-linear-agent", "canonical_url": "https://linear.app/changelog/2026-03-24-introducing-linear-agent", "kind": "release", "provenance_class": "first-party", "role": "evidence", "published_at": "2026-03-24", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "linear-agent-source-6", "original_url": "https://linear.app/changelog/2026-03-24-introducing-linear-agent", "final_url": "https://linear.app/changelog/2026-03-24-introducing-linear-agent", "captured_at": "2026-08-31T17:44:08Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/linear-agent-source-6/content.md", "sha256": "sha256:d6ea99898a8e8e273a691d8e03a7cf76dcbdef9ef89361a49330dcc5127c7555", "bytes": 5914 } } }, "approach_id": "linear-agent" }, { "id": "microsoft-prassistant-source-1", "title": "Enhancing code quality at scale with AI-powered code reviews", "url": "https://devblogs.microsoft.com/engineering-at-microsoft/enhancing-code-quality-at-scale-with-ai-powered-code-reviews/", "canonical_url": "https://devblogs.microsoft.com/engineering-at-microsoft/enhancing-code-quality-at-scale-with-ai-powered-code-reviews/", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "published_at": "2025-07-14", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "microsoft-prassistant-source-1", "original_url": "https://devblogs.microsoft.com/engineering-at-microsoft/enhancing-code-quality-at-scale-with-ai-powered-code-reviews/", "final_url": "https://devblogs.microsoft.com/engineering-at-microsoft/enhancing-code-quality-at-scale-with-ai-powered-code-reviews/", "captured_at": "2026-08-31T17:55:39Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/microsoft-prassistant-source-1/content.md", "sha256": "sha256:b1e5a99a03d2996ce76bf48f7720853b04e645e50ac7f7af53d14f4238f2d405", "bytes": 11219 } } }, "approach_id": "microsoft-prassistant" }, { "id": "monday-sphera-atlas-morphex-source-1", "title": "AI Teammates: how monday.com runs production AI agents on Amazon Bedrock", "url": "https://aws.amazon.com/blogs/machine-learning/ai-teammates-how-monday-com-runs-production-ai-agents-on-amazon-bedrock/", "canonical_url": "https://aws.amazon.com/blogs/machine-learning/ai-teammates-how-monday-com-runs-production-ai-agents-on-amazon-bedrock/", "kind": "case-study", "provenance_class": "independent-secondary", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "monday-sphera-atlas-morphex-source-1", "original_url": "https://aws.amazon.com/blogs/machine-learning/ai-teammates-how-monday-com-runs-production-ai-agents-on-amazon-bedrock/", "final_url": "https://aws.amazon.com/blogs/machine-learning/ai-teammates-how-monday-com-runs-production-ai-agents-on-amazon-bedrock/", "captured_at": "2026-08-31T17:44:51Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/monday-sphera-atlas-morphex-source-1/content.md", "sha256": "sha256:25216f6eb831f212852e1852a69cd661ce1c97e998a774dd2ed9acbaf019d9ab", "bytes": 18211 } } }, "role": "evidence", "approach_id": "monday-sphera-atlas-morphex" }, { "id": "notion-custom-agents-source-1", "title": "Notion's Token Town: 5 Rebuilds, 100+ Tools (Latent Space)", "url": "https://latent.space/p/notion", "canonical_url": "https://latent.space/p/notion", "kind": "podcast", "provenance_class": "direct-participant", "role": "evidence", "published_at": "2026-04-14", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "notion-custom-agents-source-1", "original_url": "https://latent.space/p/notion", "final_url": "https://www.latent.space/p/notion", "captured_at": "2026-08-31T17:45:03Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/notion-custom-agents-source-1/content.md", "sha256": "sha256:4e42077011382b9f77410822df8528a03212a6ed34d17c92c86c4da95acb0c89", "bytes": 124631 } } }, "approach_id": "notion-custom-agents" }, { "id": "notion-custom-agents-source-2", "title": "How we built security into Custom Agents", "url": "https://www.notion.com/en-gb/blog/how-we-built-security-into-custom-agents", "canonical_url": "https://www.notion.com/en-gb/blog/how-we-built-security-into-custom-agents", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "notion-custom-agents-source-2", "original_url": "https://www.notion.com/en-gb/blog/how-we-built-security-into-custom-agents", "final_url": "https://www.notion.com/en-gb/blog/how-we-built-security-into-custom-agents", "captured_at": "2026-08-31T17:45:14Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/notion-custom-agents-source-2/content.md", "sha256": "sha256:c442cf76acff653271f723f00e785e6ed3c069b3136efe38f35439e706742a3c", "bytes": 7751 } } }, "approach_id": "notion-custom-agents" }, { "id": "plaid-ai-annotator-source-1", "title": "AI agents at Plaid (June 2025)", "url": "https://plaid.com/blog/ai-agents-june-2025/", "canonical_url": "https://plaid.com/blog/ai-agents-june-2025/", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "published_at": "2025-06", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "plaid-ai-annotator-source-1", "original_url": "https://plaid.com/blog/ai-agents-june-2025/", "final_url": "https://plaid.com/blog/ai-agents-june-2025/", "captured_at": "2026-08-31T17:45:23Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/plaid-ai-annotator-source-1/content.md", "sha256": "sha256:452c6221596bb7d5dcb8392e5197623483ab51c606345a41d6fae741cb051c30", "bytes": 5190 } } }, "approach_id": "plaid-ai-annotator" }, { "id": "plaid-fix-my-connection-source-1", "title": "AI agents at Plaid (June 2025)", "url": "https://plaid.com/blog/ai-agents-june-2025/", "canonical_url": "https://plaid.com/blog/ai-agents-june-2025/", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "published_at": "2025-06", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "plaid-fix-my-connection-source-1", "original_url": "https://plaid.com/blog/ai-agents-june-2025/", "final_url": "https://plaid.com/blog/ai-agents-june-2025/", "captured_at": "2026-08-31T17:45:33Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/plaid-fix-my-connection-source-1/content.md", "sha256": "sha256:7fc4e30083766bf66d62e92ee0af10c514e09a0d0b6341422644017d14ab7857", "bytes": 5195 } } }, "approach_id": "plaid-fix-my-connection" }, { "id": "plaid-internal-mcp-server-source-1", "title": "The Plaid internal MCP server", "url": "https://engineering.plaid.com/the-plaid-internal-mcp-server-8eff08bb6bdb", "canonical_url": "https://engineering.plaid.com/the-plaid-internal-mcp-server-8eff08bb6bdb", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "plaid-internal-mcp-server-source-1", "original_url": "https://engineering.plaid.com/the-plaid-internal-mcp-server-8eff08bb6bdb", "final_url": "https://engineering.plaid.com/the-plaid-internal-mcp-server-8eff08bb6bdb", "captured_at": "2026-08-31T17:45:43Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/plaid-internal-mcp-server-source-1/content.md", "sha256": "sha256:eaef686c595e813529b0393b47da2d86d96e3508ebf5e8641f548bdece56a1c1", "bytes": 11749 } } }, "approach_id": "plaid-internal-mcp-server" }, { "id": "posthog-stamphog-source-1", "title": "Stop being the code review bottleneck", "url": "https://posthog.com/newsletter/code-review-tips", "canonical_url": "https://posthog.com/newsletter/code-review-tips", "kind": "corporate-article", "provenance_class": "first-party", "role": "evidence", "published_at": "2026-07-09", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "posthog-stamphog-source-1", "original_url": "https://posthog.com/newsletter/code-review-tips", "final_url": "https://posthog.com/newsletter/code-review-tips", "captured_at": "2026-08-31T17:45:51Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/posthog-stamphog-source-1/content.md", "sha256": "sha256:4cc3f02d4188cf9d525aa29d16f7e578553c4acdb180eb43d2db3a5382cf72ad", "bytes": 15950 } } }, "approach_id": "posthog-stamphog" }, { "id": "posthog-stamphog-source-2", "title": "10,000 PRs a month is easy: How devex is evolving at PostHog", "url": "https://posthog.com/blog/10k-prs-a-month", "canonical_url": "https://posthog.com/blog/10k-prs-a-month", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "authors": [ "Paul D'Ambra" ], "published_at": "2026-07-28", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "posthog-stamphog-source-2", "original_url": "https://posthog.com/blog/10k-prs-a-month", "final_url": "https://posthog.com/blog/10k-prs-a-month", "captured_at": "2026-08-31T17:46:04Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/posthog-stamphog-source-2/content.md", "sha256": "sha256:cd5dcde3b2e17a10a8d5ac8de33dbc3ef9ae92ec6cb438163dbe9c9d8fe4d79c", "bytes": 14035 } } }, "approach_id": "posthog-stamphog" }, { "id": "posthog-stamphog-source-3", "title": "StampHog PR approval agent source code and documentation", "url": "https://github.com/PostHog/posthog/tree/988c9031bb93c74bafcdfb670c01497c79a4f644/tools/pr-approval-agent", "canonical_url": "https://github.com/PostHog/posthog/tree/988c9031bb93c74bafcdfb670c01497c79a4f644/tools/pr-approval-agent", "kind": "source-code", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "posthog-stamphog-source-3", "original_url": "https://github.com/PostHog/posthog/tree/988c9031bb93c74bafcdfb670c01497c79a4f644/tools/pr-approval-agent", "final_url": "https://github.com/PostHog/posthog/tree/988c9031bb93c74bafcdfb670c01497c79a4f644/tools/pr-approval-agent", "captured_at": "2026-08-31T17:46:13Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/posthog-stamphog-source-3/content.md", "sha256": "sha256:7f4e1844030a44ed0120e3de99db3e847390fb1ee7658779d5b7908608c8bc1e", "bytes": 15840 } } }, "approach_id": "posthog-stamphog" }, { "id": "ramp-inspect-source-1", "title": "Why We Built Our Own Background Agent (Inspect)", "url": "https://engineering.ramp.com/post/why-we-built-our-background-agent", "canonical_url": "https://engineering.ramp.com/post/why-we-built-our-background-agent", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "ramp-inspect-source-1", "original_url": "https://engineering.ramp.com/post/why-we-built-our-background-agent", "final_url": "https://engineering.ramp.com/post/why-we-built-our-background-agent", "captured_at": "2026-08-31T17:46:23Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/ramp-inspect-source-1/content.md", "sha256": "sha256:4d2a6cf3edaf91cd85f6025c726ceef2b425ade364dd09a9c141d927e696d1c1", "bytes": 19819 } } }, "role": "evidence", "approach_id": "ramp-inspect" }, { "id": "ramp-inspect-source-2", "title": "Ramp x Linear (75% of merged PRs via Inspect)", "url": "https://linear.app/customers/ramp", "canonical_url": "https://linear.app/customers/ramp", "kind": "case-study", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "ramp-inspect-source-2", "original_url": "https://linear.app/customers/ramp", "final_url": "https://linear.app/customers/ramp", "captured_at": "2026-08-31T17:46:31Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/ramp-inspect-source-2/content.md", "sha256": "sha256:124a2991f6eb924e9598018e2302d7116d7125713ac365fd1e8ebd56cfb6f4a9", "bytes": 3580 } } }, "role": "evidence", "approach_id": "ramp-inspect" }, { "id": "ramp-inspect-source-3", "title": "Hacker News project discussion inspired by Ramp Inspect", "url": "https://news.ycombinator.com/item?id=48042123", "canonical_url": "https://news.ycombinator.com/item?id=48042123", "kind": "hn-thread", "provenance_class": "community", "role": "discovery", "published_at": "2026-05", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "ramp-inspect-source-3", "original_url": "https://news.ycombinator.com/item?id=48042123", "final_url": "https://news.ycombinator.com/item?id=48042123", "captured_at": "2026-08-31T17:46:39Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/ramp-inspect-source-3/content.md", "sha256": "sha256:02c79b3e6035055497be19ee0ce8b9e2b2383390e3228f4cc334903913975500", "bytes": 1360 } } }, "approach_id": "ramp-inspect" }, { "id": "ramp-inspect-source-4", "title": "Why We Built Our Own Background Agent (Inspect) (Ramp Builders URL)", "url": "https://builders.ramp.com/post/why-we-built-our-background-agent", "canonical_url": "https://builders.ramp.com/post/why-we-built-our-background-agent", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "duplicate_of": "ramp-inspect-source-1", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "ramp-inspect-source-4", "original_url": "https://builders.ramp.com/post/why-we-built-our-background-agent", "final_url": "https://builders.ramp.com/post/why-we-built-our-background-agent", "captured_at": "2026-08-31T17:46:48Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/ramp-inspect-source-4/content.md", "sha256": "sha256:50c3360bce619d6fceedd30720ba6fd8f85748f3efcd4424d0875b471792d5b9", "bytes": 19814 } } }, "approach_id": "ramp-inspect" }, { "id": "ramp-inspect-source-5", "title": "Why Ramp built its own in-house coding agent, Inspect", "url": "https://newsletter.pragmaticengineer.com/p/why-ramp-built-inspect", "canonical_url": "https://newsletter.pragmaticengineer.com/p/why-ramp-built-inspect", "kind": "news", "provenance_class": "independent-secondary", "publisher": "The Pragmatic Engineer", "authors": [ "Gergely Orosz", "Jessica Salmon", "Ivan Klaric" ], "published_at": "2026-08-25", "accessed_at": "2026-08-26", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "ramp-inspect-source-5", "original_url": "https://newsletter.pragmaticengineer.com/p/why-ramp-built-inspect", "final_url": "https://newsletter.pragmaticengineer.com/p/why-ramp-built-inspect", "captured_at": "2026-08-31T17:46:58Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/ramp-inspect-source-5/content.md", "sha256": "sha256:9cd01f7e55bbca8a2bf939c029b5ee976dbe65f84959a141c717214109bbc83f", "bytes": 15151 } } }, "role": "evidence", "approach_id": "ramp-inspect" }, { "id": "replit-manager-agent-source-1", "title": "The Self-Driving Company", "url": "https://ld.replit.com/blog/self-driving-company", "canonical_url": "https://ld.replit.com/blog/self-driving-company", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "replit-manager-agent-source-1", "original_url": "https://ld.replit.com/blog/self-driving-company", "final_url": "https://ld.replit.com/blog/self-driving-company", "captured_at": "2026-08-31T17:55:49Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/replit-manager-agent-source-1/content.md", "sha256": "sha256:5574ad7e12c3c3ffaa8b6a892da4dec151f8c7a9f48ee766f508b8d583a31eb2", "bytes": 15860 } } }, "role": "evidence", "approach_id": "replit-manager-agent" }, { "id": "retool-retoolgpt-source-1", "title": "How we built RetoolGPT", "url": "https://retool.com/blog/how-we-built-retoolgpt", "canonical_url": "https://retool.com/blog/how-we-built-retoolgpt", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "published_at": "2025-08-05", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "retool-retoolgpt-source-1", "original_url": "https://retool.com/blog/how-we-built-retoolgpt", "final_url": "https://retool.com/blog/how-we-built-retoolgpt", "captured_at": "2026-08-31T17:47:13Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/retool-retoolgpt-source-1/content.md", "sha256": "sha256:ea0306fa05f826a3864b8fb86a421c31981b363440df166c5d88004a6ec27a97", "bytes": 16758 } } }, "approach_id": "retool-retoolgpt" }, { "id": "retool-retoolgpt-source-2", "title": "AI Build Week, Day 3: How we made RetoolGPT", "url": "https://www.youtube.com/watch?v=8VTdYUBAZsY", "canonical_url": "https://www.youtube.com/watch?v=8VTdYUBAZsY", "kind": "talk", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "retool-retoolgpt-source-2", "original_url": "https://www.youtube.com/watch?v=8VTdYUBAZsY", "final_url": "https://www.youtube.com/watch?v=8VTdYUBAZsY", "captured_at": "2026-08-31T17:47:24Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/retool-retoolgpt-source-2/content.md", "sha256": "sha256:abb7d3c8ff3c0d99dfbf2d60f87d3951bdc8cdcc67e6679bb3e6dacc541db926", "bytes": 1353 } } }, "approach_id": "retool-retoolgpt" }, { "id": "salesforce-slackbot-source-1", "title": "Salesforce announces general availability of Slackbot", "url": "https://www.salesforce.com/ap/news/press-releases/2026/01/14/salesforce-announces-the-general-availability-of-slackbot-your-personal-agent-for-work-sg/", "canonical_url": "https://www.salesforce.com/ap/news/press-releases/2026/01/14/salesforce-announces-the-general-availability-of-slackbot-your-personal-agent-for-work-sg/", "kind": "release", "provenance_class": "first-party", "published_at": "2026-01-14", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "salesforce-slackbot-source-1", "original_url": "https://www.salesforce.com/ap/news/press-releases/2026/01/14/salesforce-announces-the-general-availability-of-slackbot-your-personal-agent-for-work-sg/", "final_url": "https://www.salesforce.com/ap/news/press-releases/2026/01/14/salesforce-announces-the-general-availability-of-slackbot-your-personal-agent-for-work-sg/", "captured_at": "2026-08-31T17:47:34Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/salesforce-slackbot-source-1/content.md", "sha256": "sha256:c14fe270bc355074a54043066df3b7dbbc7217db4c3ffc52ad1c5317cbe40bc6", "bytes": 13509 } } }, "role": "evidence", "approach_id": "salesforce-slackbot" }, { "id": "salesforce-slackbot-source-2", "title": "Interview with Slackbot about connected work context", "url": "https://www.salesforce.com/in/news/stories/salesforce-slackbot-ai-interview/", "canonical_url": "https://www.salesforce.com/in/news/stories/salesforce-slackbot-ai-interview/", "kind": "corporate-article", "provenance_class": "first-party", "role": "evidence", "published_at": "2025-12-01", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "salesforce-slackbot-source-2", "original_url": "https://www.salesforce.com/in/news/stories/salesforce-slackbot-ai-interview/", "final_url": "https://www.salesforce.com/in/news/stories/salesforce-slackbot-ai-interview/", "captured_at": "2026-08-31T17:47:44Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/salesforce-slackbot-source-2/content.md", "sha256": "sha256:66285b2f071812a5051b7c8470c0218f48bb712235ec57de9c345c101e8bef72", "bytes": 9504 } } }, "approach_id": "salesforce-slackbot" }, { "id": "salesforce-slackbot-source-3", "title": "Salesforce rolls out new Slackbot AI agent", "url": "https://venturebeat.com/technology/salesforce-rolls-out-new-slackbot-ai-agent-as-it-battles-microsoft-and", "canonical_url": "https://venturebeat.com/technology/salesforce-rolls-out-new-slackbot-ai-agent-as-it-battles-microsoft-and", "kind": "news", "provenance_class": "independent-secondary", "role": "commentary", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "salesforce-slackbot-source-3", "original_url": "https://venturebeat.com/technology/salesforce-rolls-out-new-slackbot-ai-agent-as-it-battles-microsoft-and", "final_url": "https://venturebeat.com/technology/salesforce-rolls-out-new-slackbot-ai-agent-as-it-battles-microsoft-and", "captured_at": "2026-08-31T17:47:51Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/salesforce-slackbot-source-3/content.md", "sha256": "sha256:e2466be395e0da687d028e1702da194650ab68f1992652e9226f9987e348f483", "bytes": 17732 } } }, "approach_id": "salesforce-slackbot" }, { "id": "salesforce-slackbot-source-4", "title": "Hacker News submission for the Salesforce Slackbot rollout", "url": "https://news.ycombinator.com/item?id=46600760", "canonical_url": "https://news.ycombinator.com/item?id=46600760", "kind": "hn-thread", "provenance_class": "community", "role": "discovery", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "salesforce-slackbot-source-4", "original_url": "https://news.ycombinator.com/item?id=46600760", "final_url": "https://news.ycombinator.com/item?id=46600760", "captured_at": "2026-08-31T17:47:58Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/salesforce-slackbot-source-4/content.md", "sha256": "sha256:e53a0fb8b2084605b1e994de831d053b1fcdffacd4282744cd0f9545fbfce238", "bytes": 566 } } }, "approach_id": "salesforce-slackbot" }, { "id": "sentry-junior-source-1", "title": "Building an Intern (Junior at Sentry)", "url": "https://cra.mr/building-an-intern/", "canonical_url": "https://cra.mr/building-an-intern/", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "sentry-junior-source-1", "original_url": "https://cra.mr/building-an-intern/", "final_url": "https://cra.mr/building-an-intern/", "captured_at": "2026-08-31T17:48:07Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/sentry-junior-source-1/content.md", "sha256": "sha256:51fee97548429baa049c0faf6a90d93bcc4891bdea99eac6cfa94fe0b003b9cc", "bytes": 28000 } } }, "role": "evidence", "approach_id": "sentry-junior" }, { "id": "sentry-junior-source-2", "title": "getsentry/junior source repository", "url": "https://github.com/getsentry/junior", "canonical_url": "https://github.com/getsentry/junior", "kind": "repository", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "sentry-junior-source-2", "original_url": "https://github.com/getsentry/junior", "final_url": "https://github.com/getsentry/junior", "captured_at": "2026-08-31T17:48:16Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/sentry-junior-source-2/content.md", "sha256": "sha256:7f3d4d7f05a8309d287a87bb2a7bf9f0018e40d83cc80b13f0d5a9adde3f9dbf", "bytes": 2512 } } }, "role": "evidence", "approach_id": "sentry-junior" }, { "id": "sentry-junior-source-3", "title": "Sentry Labs", "url": "https://labs.sentry.dev/", "canonical_url": "https://labs.sentry.dev/", "kind": "documentation", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "sentry-junior-source-3", "original_url": "https://labs.sentry.dev/", "final_url": "https://labs.sentry.dev/", "captured_at": "2026-08-31T17:48:27Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/sentry-junior-source-3/content.md", "sha256": "sha256:b81bca2f99265e347a968ce4976c507cd2b7bfd747e057f5fc2204123d5198f0", "bytes": 2439 } } }, "role": "evidence", "approach_id": "sentry-junior" }, { "id": "sentry-junior-source-4", "title": "getsentry/junior Apache 2.0 license", "url": "https://github.com/getsentry/junior/blob/main/LICENSE?plain=1", "canonical_url": "https://github.com/getsentry/junior/blob/main/LICENSE?plain=1", "kind": "source-code", "provenance_class": "first-party", "role": "evidence", "accessed_at": "2026-08-31", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "sentry-junior-source-4", "original_url": "https://github.com/getsentry/junior/blob/main/LICENSE?plain=1", "final_url": "https://github.com/getsentry/junior/blob/main/LICENSE?plain=1", "captured_at": "2026-08-31T18:13:51Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/sentry-junior-source-4/content.md", "sha256": "sha256:dc33599d579924aae2f2d107d9585af5f2efeee4d22e22631ae510b7f4535faf", "bytes": 10674 } } }, "approach_id": "sentry-junior" }, { "id": "shopify-internal-agents-source-1", "title": "Under the River", "url": "https://shopify.engineering/under-the-river", "canonical_url": "https://shopify.engineering/under-the-river", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "shopify-internal-agents-source-1", "original_url": "https://shopify.engineering/under-the-river", "final_url": "https://shopify.engineering/under-the-river", "captured_at": "2026-08-31T17:48:37Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/shopify-internal-agents-source-1/content.md", "sha256": "sha256:b6959f35653333a4b1d0c121de670b362420c92d4481bfc922a6be83a0f2e848", "bytes": 13739 } } }, "role": "evidence", "approach_id": "shopify-internal-agents" }, { "id": "sierra-pinecone-source-1", "title": "Pinecone: harnessing the wisdom of the workforce", "url": "https://sierra.ai/jp/blog/pinecone-harnessing-the-wisdom-of-the-workforce", "canonical_url": "https://sierra.ai/jp/blog/pinecone-harnessing-the-wisdom-of-the-workforce", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "sierra-pinecone-source-1", "original_url": "https://sierra.ai/jp/blog/pinecone-harnessing-the-wisdom-of-the-workforce", "final_url": "https://sierra.ai/jp/blog/pinecone-harnessing-the-wisdom-of-the-workforce", "captured_at": "2026-08-31T17:48:45Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/sierra-pinecone-source-1/content.md", "sha256": "sha256:24399215c37e3bd845e9c7acd34429c292e571355757c7cb9d1bab7ef92b7343", "bytes": 16372 } } }, "role": "evidence", "approach_id": "sierra-pinecone" }, { "id": "sierra-pinecone-source-2", "title": "Building Sierra's MCP Gateway", "url": "https://sierra.ai/blog/building-sierras-mcp-gateway-an-engineering-iceberg", "canonical_url": "https://sierra.ai/blog/building-sierras-mcp-gateway-an-engineering-iceberg", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "sierra-pinecone-source-2", "original_url": "https://sierra.ai/blog/building-sierras-mcp-gateway-an-engineering-iceberg", "final_url": "https://sierra.ai/blog/building-sierras-mcp-gateway-an-engineering-iceberg", "captured_at": "2026-08-31T17:48:54Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/sierra-pinecone-source-2/content.md", "sha256": "sha256:279856d3baa539b0050f39763f27f2a01da1e2c0a82630ace0070b6940c459b2", "bytes": 14727 } } }, "role": "evidence", "approach_id": "sierra-pinecone" }, { "id": "sierra-pinecone-source-3", "title": "Agency: secure, scalable sandboxes for agents", "url": "https://sierra.ai/es/blog/agency-secure-scalable-sandboxes-for-agents", "canonical_url": "https://sierra.ai/es/blog/agency-secure-scalable-sandboxes-for-agents", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "sierra-pinecone-source-3", "original_url": "https://sierra.ai/es/blog/agency-secure-scalable-sandboxes-for-agents", "final_url": "https://sierra.ai/es/blog/agency-secure-scalable-sandboxes-for-agents", "captured_at": "2026-08-31T17:49:02Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/sierra-pinecone-source-3/content.md", "sha256": "sha256:6b8d53460e96c2d39444c1cda5a385e1b7b3c5a34a6e8d76030e9de1ae7e790e", "bytes": 12903 } } }, "role": "evidence", "approach_id": "sierra-pinecone" }, { "id": "slack-context-system-source-1", "title": "Managing context in long-running agentic applications", "url": "https://slack.engineering/managing-context-in-long-run-agentic-applications/", "canonical_url": "https://slack.engineering/managing-context-in-long-run-agentic-applications/", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "slack-context-system-source-1", "original_url": "https://slack.engineering/managing-context-in-long-run-agentic-applications/", "final_url": "https://slack.engineering/managing-context-in-long-run-agentic-applications/", "captured_at": "2026-08-31T17:49:11Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/slack-context-system-source-1/content.md", "sha256": "sha256:7a971b0fb7f055b66353ef8efe21fb8658f63414ebad33e935c3bd1bc12e7ae9", "bytes": 22513 } } }, "role": "evidence", "approach_id": "slack-context-system" }, { "id": "slack-context-system-source-2", "title": "How Slack manages context in long-running multi-agent systems", "url": "https://www.infoq.com/news/2026/04/slack-agent-context-management/", "canonical_url": "https://www.infoq.com/news/2026/04/slack-agent-context-management/", "kind": "news", "provenance_class": "independent-secondary", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "slack-context-system-source-2", "original_url": "https://www.infoq.com/news/2026/04/slack-agent-context-management/", "final_url": "https://www.infoq.com/news/2026/04/slack-agent-context-management/", "captured_at": "2026-08-31T17:49:20Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/slack-context-system-source-2/content.md", "sha256": "sha256:80bd0ce26c165f3c8faf87d85c517048d8468d095ef1f3c591ab77b83994327a", "bytes": 4131 } } }, "role": "evidence", "approach_id": "slack-context-system" }, { "id": "spotify-honk-xirp-source-1", "title": "Spotify's Journey with Our Background Coding Agent (Honk)", "url": "https://engineering.atspotify.com/2025/11/spotifys-background-coding-agent-part-1", "canonical_url": "https://engineering.atspotify.com/2025/11/spotifys-background-coding-agent-part-1", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "spotify-honk-xirp-source-1", "original_url": "https://engineering.atspotify.com/2025/11/spotifys-background-coding-agent-part-1", "final_url": "https://engineering.atspotify.com/2025/11/spotifys-background-coding-agent-part-1", "captured_at": "2026-08-31T17:49:28Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/spotify-honk-xirp-source-1/content.md", "sha256": "sha256:6c12849337e09e69062f73492c7c8a622b4aafe43f86bfe742b08d2fd2bae2e9", "bytes": 11406 } } }, "role": "evidence", "approach_id": "spotify-honk-xirp" }, { "id": "spotify-honk-xirp-source-2", "title": "Code with Claude: coding is no longer the constraint (Honk v2)", "url": "https://engineering.atspotify.com/2026/6/code-with-claude-coding-is-no-longer-the-constraint", "canonical_url": "https://engineering.atspotify.com/2026/6/code-with-claude-coding-is-no-longer-the-constraint", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "spotify-honk-xirp-source-2", "original_url": "https://engineering.atspotify.com/2026/6/code-with-claude-coding-is-no-longer-the-constraint", "final_url": "https://engineering.atspotify.com/2026/6/code-with-claude-coding-is-no-longer-the-constraint", "captured_at": "2026-08-31T17:49:36Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/spotify-honk-xirp-source-2/content.md", "sha256": "sha256:7f48ce4eb8800a4c556d293c56b18bd6c118610b6271e6f09b2d12ea9fbc6aa9", "bytes": 11785 } } }, "role": "evidence", "approach_id": "spotify-honk-xirp" }, { "id": "spotify-honk-xirp-source-3", "title": "Xirp - Powered by Spotify Portal", "url": "https://xirp.spotify.com/", "canonical_url": "https://xirp.spotify.com/", "kind": "documentation", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "spotify-honk-xirp-source-3", "original_url": "https://xirp.spotify.com/", "final_url": "https://xirp.spotify.com/", "captured_at": "2026-08-31T17:49:43Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/spotify-honk-xirp-source-3/content.md", "sha256": "sha256:293357961af6ee356d413812d6531198688d200c2499cbac7268864bda18f9ea", "bytes": 3565 } } }, "role": "evidence", "approach_id": "spotify-honk-xirp" }, { "id": "stripe-minions-source-1", "title": "Minions: Stripe's one-shot end-to-end coding agents", "url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents", "canonical_url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "stripe-minions-source-1", "original_url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents", "final_url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents", "captured_at": "2026-08-31T17:49:52Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/stripe-minions-source-1/content.md", "sha256": "sha256:739a95644236ad907e123e95dbf38e5cb0baca359bca73a9ce1a666ad3aa915a", "bytes": 9806 } } }, "role": "evidence", "approach_id": "stripe-minions" }, { "id": "stripe-minions-source-2", "title": "Minions - Part Two", "url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents-part-2", "canonical_url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents-part-2", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "stripe-minions-source-2", "original_url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents-part-2", "final_url": "https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents-part-2", "captured_at": "2026-08-31T17:49:59Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/stripe-minions-source-2/content.md", "sha256": "sha256:bf3877ca588dd8fc63c3a3e521d93b98852de214beecea6ec6b061e74dbd24a4", "bytes": 15569 } } }, "role": "evidence", "approach_id": "stripe-minions" }, { "id": "stripe-minions-source-3", "title": "Hacker News discussion of Minions part two", "url": "https://news.ycombinator.com/item?id=47086557", "canonical_url": "https://news.ycombinator.com/item?id=47086557", "kind": "hn-thread", "provenance_class": "community", "role": "commentary", "authors": [ "ludovicianul" ], "published_at": "2026-02-20", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "stripe-minions-source-3", "original_url": "https://news.ycombinator.com/item?id=47086557", "final_url": "https://news.ycombinator.com/item?id=47086557", "captured_at": "2026-08-31T17:50:06Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/stripe-minions-source-3/content.md", "sha256": "sha256:a4fae36da63e8bc8ae62adc7337948c75bf9a3a7f6c1a891096d5601c9cf1c5f", "bytes": 23956 } } }, "approach_id": "stripe-minions" }, { "id": "stripe-minions-source-4", "title": "Hacker News comment asking for examples and review evidence", "url": "https://news.ycombinator.com/item?id=47086907", "canonical_url": "https://news.ycombinator.com/item?id=47086907", "kind": "hn-comment", "provenance_class": "community", "role": "commentary", "authors": [ "3rodents" ], "published_at": "2026-02-20", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "stripe-minions-source-4", "original_url": "https://news.ycombinator.com/item?id=47086907", "final_url": "https://news.ycombinator.com/item?id=47086907", "captured_at": "2026-08-31T17:50:17Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/stripe-minions-source-4/content.md", "sha256": "sha256:c1fee30ef91d8ae067c9a71ed3acccd9172a9e5a6a03d3e988300f266c5fde14", "bytes": 2084 } } }, "approach_id": "stripe-minions" }, { "id": "stripe-minions-source-5", "title": "Hacker News comment questioning the level of technical detail", "url": "https://news.ycombinator.com/item?id=47086953", "canonical_url": "https://news.ycombinator.com/item?id=47086953", "kind": "hn-comment", "provenance_class": "community", "role": "commentary", "authors": [ "gas9S9zw3P9c" ], "published_at": "2026-02-20", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "stripe-minions-source-5", "original_url": "https://news.ycombinator.com/item?id=47086953", "final_url": "https://news.ycombinator.com/item?id=47086953", "captured_at": "2026-08-31T17:50:25Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/stripe-minions-source-5/content.md", "sha256": "sha256:c1c33299734afe2662126e1bd16a127f59081742328afdbfc6017b3a24fa4d75", "bytes": 992 } } }, "approach_id": "stripe-minions" }, { "id": "stripe-minions-source-6", "title": "Hacker News comment asking about shipped and maintained output", "url": "https://news.ycombinator.com/item?id=47087114", "canonical_url": "https://news.ycombinator.com/item?id=47087114", "kind": "hn-comment", "provenance_class": "community", "role": "commentary", "authors": [ "rco8786" ], "published_at": "2026-02-20", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "stripe-minions-source-6", "original_url": "https://news.ycombinator.com/item?id=47087114", "final_url": "https://news.ycombinator.com/item?id=47087114", "captured_at": "2026-08-31T17:50:32Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/stripe-minions-source-6/content.md", "sha256": "sha256:1e6f8ad9a7372e11d913ce2fa5264005bd58489a1a3df8018d3580bac593fc3c", "bytes": 2331 } } }, "approach_id": "stripe-minions" }, { "id": "stripe-minions-source-7", "title": "Hacker News comment questioning the value of pull request counts", "url": "https://news.ycombinator.com/item?id=47111453", "canonical_url": "https://news.ycombinator.com/item?id=47111453", "kind": "hn-comment", "provenance_class": "community", "role": "commentary", "authors": [ "iepathos" ], "published_at": "2026-02-22", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "stripe-minions-source-7", "original_url": "https://news.ycombinator.com/item?id=47111453", "final_url": "https://news.ycombinator.com/item?id=47111453", "captured_at": "2026-08-31T17:50:40Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/stripe-minions-source-7/content.md", "sha256": "sha256:7cca401d3c991caab60e07628506a8515859e34266d72fbe29da405add95a79e", "bytes": 1562 } } }, "approach_id": "stripe-minions" }, { "id": "uber-coding-agent-source-1", "title": "Uber's CTO on AI coding agents (Business Insider)", "url": "https://www.businessinsider.com/uber-cto-ai-coding-agentic-software-engineers-2026-3", "canonical_url": "https://www.businessinsider.com/uber-cto-ai-coding-agentic-software-engineers-2026-3", "kind": "news", "provenance_class": "independent-secondary", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "uber-coding-agent-source-1", "original_url": "https://www.businessinsider.com/uber-cto-ai-coding-agentic-software-engineers-2026-3", "final_url": "https://www.businessinsider.com/uber-cto-ai-coding-agentic-software-engineers-2026-3", "captured_at": "2026-08-31T17:55:56Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/uber-coding-agent-source-1/content.md", "sha256": "sha256:9e2663eb9d3a19995388b0c38f0dbb067d7115f4677809cd1bf69237fb231a5f", "bytes": 3503 } } }, "role": "evidence", "approach_id": "uber-coding-agent" }, { "id": "uber-ureview-source-1", "title": "uReview: Scalable, Trustworthy GenAI for Code Review at Uber", "url": "https://www.uber.com/us/en/blog/ureview/", "canonical_url": "https://www.uber.com/us/en/blog/ureview/", "kind": "engineering-blog", "provenance_class": "first-party", "role": "evidence", "published_at": "2025-08-12", "accessed_at": "2026-08-13", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "uber-ureview-source-1", "original_url": "https://www.uber.com/us/en/blog/ureview/", "final_url": "https://www.uber.com/us/en/blog/ureview/", "captured_at": "2026-08-31T17:51:23Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/uber-ureview-source-1/content.md", "sha256": "sha256:c59a0d07af90fbc87d6cee3dfb9d99ead426723b4e0c8d5b16a198f756dc5a6e", "bytes": 22823 } } }, "approach_id": "uber-ureview" }, { "id": "workos-project-horizon-source-1", "title": "Project Horizon - an autonomous code factory at WorkOS", "url": "https://workos.com/blog/project-horizon", "canonical_url": "https://workos.com/blog/project-horizon", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "workos-project-horizon-source-1", "original_url": "https://workos.com/blog/project-horizon", "final_url": "https://workos.com/blog/project-horizon", "captured_at": "2026-08-31T17:51:33Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/workos-project-horizon-source-1/content.md", "sha256": "sha256:d65c744f67e8dcebff0bd7a541dc513e650d496825b2b6d7caa5fccc415bcdbf", "bytes": 22458 } } }, "role": "evidence", "approach_id": "workos-project-horizon" }, { "id": "workos-project-horizon-source-2", "title": "Applied AI Showcase (Horizon with Claude Remote Routines)", "url": "https://workos.com/blog/applied-ai-showcase", "canonical_url": "https://workos.com/blog/applied-ai-showcase", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "workos-project-horizon-source-2", "original_url": "https://workos.com/blog/applied-ai-showcase", "final_url": "https://workos.com/blog/applied-ai-showcase", "captured_at": "2026-08-31T17:51:42Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/workos-project-horizon-source-2/content.md", "sha256": "sha256:39859b8507602df7bd868c0c1ff702ab82e5b74155f18c5a5608ed06233952f0", "bytes": 10305 } } }, "role": "evidence", "approach_id": "workos-project-horizon" }, { "id": "workos-project-horizon-source-3", "title": "An autonomous UI-quality program", "url": "https://workos.com/blog/autonomous-ui-quality-program", "canonical_url": "https://workos.com/blog/autonomous-ui-quality-program", "kind": "engineering-blog", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "workos-project-horizon-source-3", "original_url": "https://workos.com/blog/autonomous-ui-quality-program", "final_url": "https://workos.com/blog/autonomous-ui-quality-program", "captured_at": "2026-08-31T17:51:49Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/workos-project-horizon-source-3/content.md", "sha256": "sha256:d6aa4db6f84168d77c3025f59a65f066a8d0e7df847aad2e3071bb1602987420", "bytes": 6360 } } }, "role": "evidence", "approach_id": "workos-project-horizon" }, { "id": "workos-project-horizon-source-4", "title": "Hacker News submission for Project Horizon", "url": "https://news.ycombinator.com/item?id=48039227", "canonical_url": "https://news.ycombinator.com/item?id=48039227", "kind": "hn-thread", "provenance_class": "community", "role": "discovery", "authors": [ "i0exception" ], "published_at": "2026-05-06", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "workos-project-horizon-source-4", "original_url": "https://news.ycombinator.com/item?id=48039227", "final_url": "https://news.ycombinator.com/item?id=48039227", "captured_at": "2026-08-31T17:51:59Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/workos-project-horizon-source-4/content.md", "sha256": "sha256:9ce203a78d2e0e764ba23813d5c5968b9deddae29a3bc6c052cb8a8e465a96f9", "bytes": 413 } } }, "approach_id": "workos-project-horizon" }, { "id": "ycombinator-agent-infra-source-1", "title": "Inside YC's AI Playbook (Lightcone podcast, with Pete Koomen)", "url": "https://www.ycombinator.com/library/Qh-inside-yc-s-ai-playbook", "canonical_url": "https://www.ycombinator.com/library/Qh-inside-yc-s-ai-playbook", "kind": "podcast", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "ycombinator-agent-infra-source-1", "original_url": "https://www.ycombinator.com/library/Qh-inside-yc-s-ai-playbook", "final_url": "https://www.ycombinator.com/library/Qh-inside-yc-s-ai-playbook", "captured_at": "2026-08-31T17:56:06Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/ycombinator-agent-infra-source-1/content.md", "sha256": "sha256:2432ab3ab9cfeb941654962c5b066980eb2f7e61aae848e8bd306d2f6fc30c4a", "bytes": 51806 } } }, "role": "evidence", "approach_id": "ycombinator-agent-infra" }, { "id": "zup-codegen-source-1", "title": "Building an Internal Coding Agent at Zup", "url": "https://arxiv.org/abs/2604.09805", "canonical_url": "https://arxiv.org/abs/2604.09805", "kind": "paper", "provenance_class": "first-party", "accessed_at": "2026-08-12", "last_verified_at": "2026-08-31", "capture": { "schema_version": 1, "source_id": "zup-codegen-source-1", "original_url": "https://arxiv.org/abs/2604.09805", "final_url": "https://arxiv.org/abs/2604.09805", "captured_at": "2026-08-31T17:52:42Z", "http_status": 200, "tool": { "name": "steel", "version": "0.4.4" }, "artifacts": { "markdown": { "path": "archive/sources/zup-codegen-source-1/content.md", "sha256": "sha256:738c9131fd4b99581f618b9dfe9f4c289f4b890db0eea36f58d200d0211f7eab", "bytes": 3944 } } }, "role": "evidence", "approach_id": "zup-codegen" } ] }