Methodology

What we include

Each case describes a system that a named organization built or adapted for its own teams. Public sources must describe its implementation or use.

We also include platforms and supporting tools that help teams build internal agents. A product name is optional. Commercial systems can qualify when sources describe the internal build or adaptation. General adoption claims are insufficient.

How we add cases

We use an agent skill to assess sources against our inclusion rules and prepare catalog changes. Automated checks validate record structure, source files, and links; they cannot confirm reported results or our interpretations.

How we use sources

Claims link to public sources. We record who published each source and separate reported claims from our own judgments. We keep conflicting reports visible.

We capture a copy of each source page with Steel, the open-source browser infrastructure for AI agents. Each capture records the time, the final URL, and the HTTP status. A capture shows what a page said when we read it. It does not confirm the claims on that page.

Company results remain self-reported unless an independent source verifies them. Confidence describes the support for a claim. Evidence strength describes source type and detail. Neither label proves that a claim is true.

For metrics, we keep the reported dates, scope, measurement method, and what the numbers count, when available. A report date does not establish the measurement period.

Unknown means that the sources do not provide an answer. It does not establish that a feature is absent.

How we assign levels

We assess specific tasks and when a person must review the work. Each level is our judgment, linked to evidence and a date. One system can have several levels for different tasks.

The levels do not rank companies or measure quality. See Definitions for the terms and framework.

How we write notes

We compare design choices across cases. Each note links to its sources and separates what teams report from our observations.

We state the limits of each comparison. A repeated choice does not prove that it works better. Our illustrations explain the ideas and identify simplifications.

What the map cannot tell you

The map covers cases with public evidence. Many concern coding and code review. Failed projects and unpublished systems may be missing.

Case counts cannot tell us how common a practice is across the industry. Results from different tasks or measurement methods may not be comparable.