AGENTS.md adoption crosses 60,000 repositories as AI coding agents standardize

AGENTS.md, the plain-text file that tells AI coding agents how a project works, has passed 60,000 open-source repository adoptions, according to the Agentic AI Foundation. The file is now supported by a growing list of tools, including Copilot, Cursor, Codex, Gemini CLI, Devin, and others, as AI-assisted coding workflows settle around a shared way to pass project-specific instructions to models.

The idea behind AGENTS.md is simple. Put a markdown file in the root of a repository, and give an AI agent the information it needs to work in that codebase without guessing. That can include build commands, test steps, formatting conventions, folder layout, branching rules, and other instructions that are specific to the project rather than the programming language itself.

The file was originally created by OpenAI and has since been moved under the governance of the Agentic AI Foundation, which operates under the Linux Foundation umbrella. That matters because standards become more useful when they are not tied to a single vendor. A common format makes it easier for multiple coding tools to read the same instructions instead of forcing each product to invent its own project-context system.

This is part of a broader shift in software development. AI coding agents are no longer just autocomplete tools sitting inside an editor. They are increasingly being used to inspect repositories, make changes, run tests, and carry out multi-step tasks, which means they need a reliable source of local knowledge before they can act with any confidence.

Without that context, an agent may still produce code, but it is more likely to miss project conventions or suggest changes that do not fit the repository’s build process. With a shared file at the repo root, maintainers have a lightweight way to describe how the project wants code to be handled. That also gives contributors a single place to look when they want to understand the project’s expectations.

Support from major tools is helping push AGENTS.md beyond an informal convention and toward a de facto standard. When products such as Copilot, Cursor, Codex, Gemini CLI, and Devin recognize the same file, teams can write guidance once and have it interpreted across different agent systems. That reduces duplication and makes it easier for open-source maintainers to keep instructions consistent as their tooling changes.

The adoption number also suggests that project maintainers are treating AI agents as part of the normal development workflow, not an experimental add-on. In practical terms, AGENTS.md sits alongside the code, the tests, and the README, but its purpose is narrower: it is for machine readers that need operational guidance before they edit files or run commands.

Because the file is markdown, it stays human-readable and easy to version-control. It can be reviewed in pull requests, updated as the project changes, and tailored to a repository without requiring special infrastructure or a new hosting system. That keeps the standard simple, which is usually what helps it spread.

The Agentic AI Foundation says AGENTS.md is now part of a broader effort to standardize how agentic tools interact with repositories, and the recent adoption milestone shows that the format is already showing up in a large number of projects.