OpenAI Co-Founds Agentic AI Foundation, Donates AGENTS.md

OpenAI is co-founding the Agentic AI Foundation (AAIF) under the Linux Foundation, alongside Anthropic and Block, with support from Google, Microsoft, AWS, Bloomberg, and Cloudflare. As part of the launch, OpenAI is donating AGENTS.md, its open format for giving AI agents project-specific instructions and context, to the new foundation.

The AAIF's stated purpose is to provide neutral stewardship for open, interoperable infrastructure as agentic AI systems move from experimentation into real-world production. Anthropic is contributing the Model Context Protocol (MCP), and Block is contributing its goose project. The foundation operates as a directed fund under the Linux Foundation, the non-profit that already stewards the Linux Kernel, Kubernetes, Node.js, and PyTorch.

The launch matters because the agent ecosystem is at an inflection point. In 2025, agentic systems — from coding assistants to workflow automation and customer service agents — began moving from prototypes into tools that handle real work in business and consumer settings. OpenAI argues that this transition will only work at scale if open standards make agents interoperable across tools and platforms.

The fragmentation problem

OpenAI's framing of the stakes is direct. Open standards, the company argues, make agents safer, easier to build, and more portable across tools and platforms — and help prevent the ecosystem from fragmenting as the category matures.

"As more agents begin handling real responsibility, the cost of fragmentation increases," OpenAI states. "Without common conventions and neutral governance, agent development risks diverging into incompatible silos that limit portability, safety, and progress."

That is the problem AAIF was created to address. The bet is that a neutral governance body — not any single vendor — is the mechanism that keeps foundational agent infrastructure from splintering the way earlier platform wars did. The Linux Foundation's decades of experience with neutral stewardship is the template here: the same structure that kept Kubernetes and the Linux Kernel vendor-neutral is now being applied to agent interoperability.

What OpenAI is bringing to the table

The AAIF announcement follows a year of open-source agentic infrastructure work from OpenAI. The company has contributed the Agents SDK, the Apps SDK, and the Agentic Commerce Protocol, along with open-source efforts including the gpt-oss models and the Codex CLI.

The adoption numbers on Codex are notable: OpenAI says Codex alone has now been used to help merge more than two million public pull requests on GitHub.

OpenAI has also positioned itself as an early adopter and core contributor to the Model Context Protocol — the same protocol Anthropic created and is now contributing to the AAIF. OpenAI incorporated MCP as the foundation for connectors and apps in ChatGPT. Just last week, OpenAI announced a collaboration with Anthropic and MCP-UI to extend the Apps SDK to all MCP developers through MCP Apps, which the company says makes it easier to build interactive, agent-powered experiences across platforms.

These efforts, OpenAI says, laid the groundwork for AAIF by demonstrating that open, interoperable infrastructure can gain real adoption across tools and communities.

AGENTS.md: from internal fix to 60,000-project standard

AGENTS.md itself began as a practical solution to a familiar problem. Codex needed a predictable way to find project-specific instructions — coding conventions, build steps, testing requirements — so agents could work safely and effectively inside a codebase.

The answer was a lightweight Markdown file designed to live alongside README.md. It gives AI coding agents a consistent source of project-specific guidance, making agent behavior predictable across different repositories, toolchains, and build systems.

What started as an internal solution resonated quickly across the ecosystem. Since its release in August 2025, AGENTS.md has been adopted by more than 60,000 open-source projects and by agent frameworks and tools including Amp, Codex, Cursor, Devin, Factory, Gemini CLI, GitHub Copilot, Jules, and VS Code. That adoption reflects growing alignment around shared, vendor-neutral conventions as agents enter production, according to OpenAI.

The number is the story here. A convention invented for one company's coding agent became a de facto industry standard in roughly a year — without any governance structure behind it. Donating it to the AAIF formalizes what the ecosystem had already decided in practice.

OpenAI lists three specific outcomes it expects from contributing AGENTS.md to the foundation:

• Developers and enterprises have a portable way to provide agents with the context they need
• The format can evolve in the open, with input from many tools and communities
• No single company controls its direction as adoption continues to grow

Why governance structure matters

The choice of the Linux Foundation as home is not incidental. AAIF operates as a directed fund under the non-profit, which OpenAI credits with "decades of experience stewarding critical open-source projects including the Linux Kernel, Kubernetes, Node.js, and PyTorch." The Linux Foundation brings what OpenAI describes as "proven expertise in neutral governance, community building, and maintaining long-term sustainability for widely adopted technologies — ensuring no single company controls the direction of foundational infrastructure."

That last point carries the most weight. The founding roster — OpenAI, Block, Anthropic, Google, Microsoft, Amazon, and Bloomberg, plus Cloudflare — includes the most direct competitors in AI. OpenAI and Anthropic are donating their respective agent-interfacing standards, AGENTS.md and MCP, to the same neutral body. Both companies have now committed their core interoperability formats to shared governance rather than proprietary control.

The AAIF is designed, per the announcement, to help agentic AI evolve transparently, collaboratively, and in the public interest through shared investment, open development, and community-led standards. The foundation explicitly invites contributors of all kinds — tool builders, researchers, enterprises, and individual developers — to shape and extend the work.

The so-what

The practical question for developers and enterprises is whether AAIF-governed standards meaningfully reduce the integration burden of building agents that work across multiple platforms, tools, and vendors. With AGENTS.md, MCP, and goose under one neutral roof — and with the major AI labs, cloud providers, and development toolmakers either co-founding or supporting the effort — the industry has, for the first time, a single governance structure covering the two most widely adopted pieces of agent interoperability infrastructure. Whether that translates into genuinely portable agents, or into a standards body whose pace lags the market it is trying to organize, will depend on how the foundation's community-led development process performs in practice.