An independent research collective studying how AI agents interact with documentation, tools, and web infrastructure. We publish reports and open specifications based on what we find, and build the tooling that measures compliance. Companies are already measuring their documentation against our specs and competing on their scores.

The collective is self-funded by its members. Individuals who find the work valuable can support it through GitHub Sponsors, but there’s no corporate sponsorship program and nobody pays for our conclusions: published findings reflect what the data shows. That independence is what makes the research credible and useful.

It means documentation that coding agents (like Claude Code, Cursor, GitHub Copilot) can effectively discover, access, and use. Our 23-check specification covers content discoverability, markdown availability, page size, content structure, URL stability, observability, and authentication. When documentation meets these checks, agents can use it reliably; when it doesn’t, agents fail in predictable and preventable ways.

All specifications, tools, and benchmarks live in the agent-ecosystem GitHub organization. Published reports are freely available at their dedicated sites, and aeshift.com carries commentary on ecosystem developments.

Everyone building or using agent tooling. Agent platform companies benefit when documentation works well with their agents. Developer tool companies benefit when they understand how agents consume their docs. IDE and coding tool companies benefit from better tool quality and interoperability standards. And developers benefit from an ecosystem that just works.

Yes. Our specifications, audit tools (afdocs, skill-validator), and research methodologies are all open source. Published research and reports are freely available. We believe ecosystem standards should be open by default.