AGENT.md

AGENT.md is a plain markdown file placed at the root of a repository that hands an AI coding agent the project context a human would otherwise have to explain: what the project is, how to build and test it, coding conventions, and any guardrails the agent should respect. It is the singular, vendor-neutral variant of the same idea expressed by AGENTS.md — one predictable file an agent reads before it starts changing your code, rather than a different dotfile per tool.

• Human-readable markdown - No new schema to learn; the file is just structured prose an agent and a developer can both follow.
• Vendor-neutral by design - Meant to work across whichever assistant a developer happens to use, instead of locking guidance to one product’s proprietary config.
• Project context in one place - Build commands, test commands, directory conventions, and do-not-touch areas live together where the agent looks first.
• Repo-local and version-controlled - Because it lives in the codebase, the guidance travels with the code and evolves through normal pull requests.

In practice AGENT.md is where API teams encode the operational knowledge an agent needs to work safely inside a repository — the build/test loop, the governance rules, the parts of the API surface that must not drift. It sits alongside the broader move toward machine-readable guidance I keep writing about, and its main friction is fragmentation: with both AGENT.md and AGENTS.md in circulation, teams have to decide which convention to publish so their agents stop guessing.