OpenAI donates AGENTS.md to the Linux Foundation's new Agentic AI Foundation

What OpenAI actually handed over

OpenAI is co-founding the Agentic AI Foundation (AAIF) under the Linux Foundation, together with Anthropic and Block, and with support from Google, Microsoft, AWS, Bloomberg, and Cloudflare. Its contribution to the foundation is AGENTS.md.

AGENTS.md is a lightweight Markdown file meant to sit alongside README.md. It gives AI coding agents a predictable place to find project-specific instructions — coding conventions, build steps, testing requirements — so an agent behaves consistently across different repositories and toolchains.

The format originated inside OpenAI as a practical fix: Codex needed a reliable way to locate per-project guidance. The company reports Codex has helped merge more than two million public pull requests on GitHub, so the problem AGENTS.md solves was one OpenAI hit at scale first.

Why a Markdown file counts as frontend infrastructure

AGENTS.md is not UI in the visual sense, but it functions as the interface layer between a codebase and the agents acting on it. It is the surface where a developer declares intent to a machine collaborator, and it is versioned and reviewed like any other project file.

OpenAI says the format has been adopted by more than 60,000 open-source projects since its August 2025 release, and lists agent tools including Amp, Cursor, Devin, Factory, Gemini CLI, GitHub Copilot, Jules, and VS Code as users. That breadth is the point: a convention only reduces friction if the tools your team already uses all read it the same way.

The announcement also ties AGENTS.md to broader frontend work. OpenAI notes it extended its Apps SDK to all MCP developers last week through MCP Apps, a collaboration with Anthropic and MCP-UI aimed at building interactive, agent-powered experiences across platforms.

The governance detail that changes the calculus

The AAIF operates as a directed fund under the Linux Foundation, the body that has stewarded the Linux Kernel, Kubernetes, Node.js, and PyTorch. OpenAI frames the goal plainly.

Contributing AGENTS.md to the AAIF ensures that... No single company controls its direction as adoption continues to grow.Montana Labs

That matters because the co-founders are otherwise direct competitors. Each is contributing a project it built: OpenAI's AGENTS.md, Anthropic's Model Context Protocol, and Block's goose. Placing all three under neutral stewardship is a bet that shared conventions beat proprietary lock-in as agents move into production work.

What this means when you build agent-facing surfaces

For teams shipping agent tooling, the practical takeaway is that AGENTS.md is now a low-risk convention to support. It is a plain Markdown file, cheap to add, and its direction is no longer subject to one vendor's roadmap changes.

The pairing of AGENTS.md with MCP and the MCP Apps effort points to a stack forming in the open: MCP for connecting agents to tools and data, MCP Apps for interactive experiences, and AGENTS.md for per-repository context. Designing around these three now is a cheaper hedge than betting on a single vendor's closed alternative and rebuilding later.

The specific implication is narrow but concrete: the file your agents read to understand a project is becoming a community-governed standard rather than an OpenAI artifact — which makes adopting it a portability decision, not a lock-in one.