What it is
The Agentic AI Foundation (AAIF) is a vendor-neutral organization, operated as a directed fund under the Linux Foundation, created to govern open technical standards for AI agents — the software systems that use large language models to take multi-step actions rather than just answer questions. It was co-founded by Anthropic, OpenAI, and Block (the payments company behind Square and the Goose agent framework), with additional financial and technical backing from Google, Microsoft, Amazon Web Services, Cloudflare, and Bloomberg.
The foundation does not build agents itself. Instead, it is a custodian: companies and projects donate specifications, code, trademarks, and domain names to it, and the Linux Foundation’s existing governance machinery takes over stewardship, with maintainers (drawn from multiple companies, not just the original donor) continuing day-to-day technical decisions. The model is the same one that put Kubernetes, Linux itself, and other widely used infrastructure outside the control of any one corporate sponsor.
At launch, three projects moved under the AAIF umbrella: Anthropic’s Model Context Protocol (MCP), a standard for connecting AI systems to external data and tools, which Anthropic says now powers more than 10,000 public servers and is supported in ChatGPT, Cursor, Gemini, Microsoft Copilot, and VS Code; OpenAI’s AGENTS.md, a lightweight markdown convention for giving coding agents project-specific instructions; and Block’s Goose, an open-source agent framework that moved to community governance.
Why it matters for AI governance and narratives
The foundation is a case study in how the leading AI labs are choosing to manage interoperability rather than compete on it — a notable departure from the more closed, platform-specific postures labs have taken on model weights and training data. The explicit framing from its founders is that agentic AI should “evolve transparently, collaboratively, and in the public interest,” which positions standards-sharing as a governance good distinct from, and arguably in tension with, each lab’s commercial interest in locking developers into its own agent ecosystem.
For the observatory’s purposes, the AAIF is a useful marker of where competitive and cooperative framing intersect. OpenAI and Anthropic are simultaneously rivals in model capability and co-stewards of the plumbing that lets agents built on either company’s models interoperate. When OpenAI donated its Open Responses API specification and Anthropic donated Agent Skills on the same day — October 22, 2026 — both companies had an incentive to frame the move as public-interest infrastructure-building rather than as a defensive response to fragmentation risk or regulatory scrutiny of walled-garden agent ecosystems. Readers should treat the foundation’s existence as evidence of a coordination strategy among dominant labs, not as a neutral technical fact free of competitive calculation.
Key facts and dates
The AAIF was established in December 2025 under the Linux Foundation, with Anthropic, OpenAI, and Block as founding co-sponsors. In its first year it has grown from three launch projects to ten, adding (as of October 22, 2026) Agent Skills (donated by Anthropic), Open Responses (donated by OpenAI), OpenEnv (contributed by Hugging Face), and OpenSandbox (contributed by Alibaba).
Agent Skills — a format in which a folder containing a SKILL.md file teaches an agent a specific task or workflow — was published by Anthropic as an open standard in December 2025 after launching in October 2025. Anthropic reports that more than 70 platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated it, with roughly 1 million skills published and nearly 280 million installations. Anthropic’s donation transfers the Agent Skills name, logo, GitHub organization, and the agentskills.io domain to the foundation; engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard going forward, with Anthropic’s David Soria Parra and Mahesh Murag named as lead maintainers. Open Responses, OpenAI’s specification for a common API across LLM providers, was donated the same day, intended to reduce the work developers face translating between vendors’ differing API conventions when switching models.
Where to learn more
- Donating the Model Context Protocol and Establishing the Agentic AI Foundation — Anthropic — primary source on the foundation’s founding, members, and launch projects
- Donating Agent Skills to the Agentic AI Foundation — Anthropic — primary source on the Agent Skills donation, adoption figures, and governance
- Block, Anthropic, and OpenAI Launch the Agentic AI Foundation — Block — co-founder’s framing of the foundation’s purpose and Goose’s role
- OpenAI donates Open Responses, its spec for a common LLM API, to the Agentic AI Foundation — The New Stack — independent coverage of the Open Responses donation