According to TechCrunch, the Linux Foundation is launching a new group called the Agentic AI Foundation (AAIF) to standardize the emerging world of AI agents. Founding members include OpenAI, Anthropic, and Block, who are each donating key technology: Anthropic is contributing its Model Context Protocol (MCP), Block is open-sourcing its Goose agent framework, and OpenAI is bringing its AGENTS.md specification. Other members like AWS, Bloomberg, Cloudflare, and Google are also joining. The immediate goal is to prevent a future of incompatible, “closed wall” proprietary AI agent stacks. The group will be funded through member dues but governed by technical steering committees to avoid any single company controlling the direction.
The plumbing problem
Here’s the thing about AI agents: they’re not just chatbots that talk. They’re systems designed to take actions—book a flight, analyze a spreadsheet, write and execute code. But for that to work, they need to connect to a million different tools, databases, and APIs. Right now, every company building agents is essentially creating its own proprietary plumbing. It’s a mess waiting to happen. You’d have OpenAI agents that can’t talk to Anthropic’s systems, or Google’s agents that need a completely custom adapter to work with your company’s internal software. The AAIF is basically an attempt to agree on standard pipe sizes and thread types before everyone builds their own custom, incompatible houses.
Why donate your secret sauce?
So why would these companies, especially rivals like OpenAI and Anthropic, give away what could be competitive infrastructure? For Block, it’s pretty pragmatic. Brad Axen from Block said open-sourcing Goose gives them a place for “other people to come help us make it better.” They get free R&D from the open-source community, and all improvements flow back to them. For Anthropic and OpenAI, it’s about influence and necessity. If MCP or AGENTS.md becomes the de facto standard, they’ve effectively set the rules of the game. David Soria Parra from Anthropic said the goal is to have “enough adoption in the world that it’s the de facto standard.” It’s a classic play: you can’t win the platform war alone, so you try to become the indispensable, neutral foundation everyone builds on.
The governance tightrope
But let’s be skeptical for a second. We’ve seen these “open” alliances before. Remember, the Linux Foundation itself hosts Kubernetes, which did become a massive standard, but it’s also home to projects that fade into obscurity. Jim Zemlin, the Foundation’s director, says funding doesn’t equal control, and roadmaps are set by technical committees. That’s the ideal. The real test is what happens when there’s a major technical disagreement. What if OpenAI wants AGENTS.md to evolve in a way that subtly advantages their models? Or what if Google pushes for a protocol change that integrates better with its own cloud services? Nick Cooper from OpenAI says he doesn’t want the standards to be “stagnant,” which is good, but evolution can also mean fragmentation.
A web for agents?
The big, ambitious vision here is to create for AI agents what HTTP and HTML created for the web: an open, interoperable system where anyone can build a tool that works with any agent. Instead of being locked into one company’s “agent stack,” you could mix and match—use Anthropic’s model with Block’s Goose framework, connected via MCP to your company’s data. That’s the dream. It would lower development time dramatically and could lead to a more innovative and competitive ecosystem. But it requires genuine collaboration from companies whose core business is, ultimately, competing with each other in the AI race. The success of this effort won’t be measured by press releases, but by whether developers actually start building on this shared plumbing instead of the proprietary pipes that the big platforms will undoubtedly still try to sell them.
