OpenAI, Anthropic, and Block have jointly founded the Agentic AI Foundation โ a neutral governance organization housed under the Linux Foundation โ with the goal of building open standards and shared infrastructure for AI agents. Each founding company is donating a core open-source project to seed the foundation: Anthropic’s Model Context Protocol (MCP), OpenAI’s AGENTS.md, and Block’s Goose agent framework.
What Each Company Is Contributing
The three founding contributions represent different layers of the agent stack:
- Anthropic’s Model Context Protocol (MCP): The connectivity standard that allows AI agents to communicate with external tools, databases, APIs, and data sources through a unified interface. MCP has already crossed 97 million installs and is supported by ChatGPT, Cursor, Gemini, VS Code, and over 10,000 active public servers. Anthropic previously donated MCP to the Linux Foundation; this co-founding formalizes a broader governance structure around it.
- OpenAI’s AGENTS.md: A specification format for defining agent behavior, capabilities, and constraints in a standardized, portable way. AGENTS.md allows developers to describe what an agent can and cannot do in a format that any compatible platform can interpret โ enabling agents to move across tools and environments without being rebuilt from scratch.
- Block’s Goose agent framework: An open-source framework for building and deploying AI agents developed by Block’s engineering teams, contributing production-grade implementation experience from one of the largest financial technology companies using AI agents at scale.
Why This Matters Beyond the Announcement
Three competing AI companies co-founding a neutral standards body is unusual โ and significant. The typical pattern in emerging technology is for companies to push proprietary standards and fight interoperability battles for years before the market forces convergence. The early formation of a neutral foundation around agent standards suggests the founding companies believe shared infrastructure will expand the agent market faster than proprietary lock-in would benefit any individual player.
MCP’s rapid adoption is the best evidence for this thesis. Since Anthropic open-sourced MCP, it has been adopted by OpenAI, Google, Microsoft, and dozens of major developer tools โ creating a de facto connectivity standard that benefits Anthropic by making Claude more broadly useful, while also benefiting every other AI provider that supports it. A rising tide of agent connectivity lifted all boats more than a proprietary standard would have.
The Linux Foundation’s Role
Housing the Agentic AI Foundation under the Linux Foundation is a deliberate choice. The Linux Foundation has a proven track record of managing neutral governance for foundational open-source technology โ Linux itself, Kubernetes, and dozens of other infrastructure projects that became industry standards. Its governance model prevents any single company from controlling the direction of shared standards, which is essential for getting competitors to contribute and adopt common specifications.
The foundation structure also provides a formal mechanism for other companies to join as contributors or members โ expanding beyond the three founding companies to include the broader developer tool ecosystem, cloud providers, and enterprise software vendors that are building on top of agent infrastructure.
What Developers Get
For developers building agents, the foundation’s output is interoperability. Today, an agent built with Anthropic’s tooling may not easily connect to tools designed for OpenAI’s ecosystem, and vice versa. Shared standards โ for how agents communicate, how they declare their capabilities, and how they connect to external tools โ reduce the rebuild cost when moving between platforms and make agent infrastructure more durable as the ecosystem evolves.
The MCP registry already lists over 4,000 published servers covering SaaS platforms, enterprise systems, development tools, and specialized data sources. AGENTS.md and Goose add specification and framework layers that, if widely adopted, could make building production agents significantly more straightforward than it is today.
Conclusion
The Agentic AI Foundation is the clearest signal yet that the foundational infrastructure layer of AI agents will be governed by open standards rather than proprietary lock-in. Whether that bet plays out depends on how broadly the ecosystem adopts AGENTS.md and the frameworks that build on it โ but the combination of OpenAI, Anthropic, and Block pooling their contributions gives the foundation the legitimacy to attract the adoption it needs. Browse our directory to explore the agent tools building on this shared infrastructure.
