You use three AI tools at work. None of them talk to each other. Every integration is duct tape and prayers. Your IT team builds custom connectors that break with every update. This is 2026, and AI agents — programs that do tasks on your behalf — still can't agree on how to shake hands.

The Linux Foundation — the neutral body that governs Linux, Kubernetes, and Node.js — stepped in. On December 9, 2025, they launched the Agentic AI Foundation (AAIF): a single organization to govern the protocols (agreed-upon rules for communication) that make AI agents actually work together.

Three founding projects came with it:

  • MCP (Model Context Protocol) from Anthropic — a universal plug standard for connecting AI to tools and data. Think USB-C, but for AI integrations.
  • goose from Block — an open-source (code anyone can inspect and modify) agent framework, basically a skeleton for building AI agents that use tools, running locally on your machine.
  • AGENTS.md from OpenAI — a simple text file that tells an AI agent what a project is and what rules to follow. Boring name. Brilliant idea.

Platinum members at the table: AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI. Every major AI company, in one room, agreeing on something. Mark your calendar — this doesn't happen often.

On February 24, 2026, AAIF announced 97 new members — 18 Gold, 79 Silver — bringing total membership to 146. David Nalley chairs the governing board. The MCP Dev Summit North America runs April 2–3 in New York.

Why does this matter to you?

Standards die without governance — a neutral body that decides how the standard evolves so no single company controls it. MCP was already winning under Anthropic's roof, but enterprises quietly asked the same question: what if Anthropic changes the rules to favor their own AI, Claude? What if they pivot and leave everyone stranded?

Moving MCP to the Linux Foundation kills that fear. No single company can unilaterally change the spec (the technical specification that defines how MCP works). "Anthropic's protocol" always faces competitor resistance. "The Linux Foundation's protocol" faces none.

Anthropic traded control for universal adoption. That's not charity — that's a grandmaster move. They built the standard, let it win, then made it everyone's problem to maintain. The protocol still carries Anthropic's DNA. The governance just has a neutral address.

And OpenAI? In March 2025, they launched the Agents SDK and Responses API — their own framework with proprietary tool-calling, widely seen as a direct alternative to MCP. By June 2025, they reversed course: Sam Altman announced MCP support across OpenAI products. Now they're co-founding the foundation that governs MCP and contributing AGENTS.md to the same umbrella. Translation: the protocol war is over. MCP won. OpenAI pivoted from "build our own" to "influence the winner from inside."

The stack

These three projects aren't competing — they're layers:

  • MCP = how agents talk to tools (the protocol)
  • goose = how you build an agent that uses those tools (the framework)
  • AGENTS.md = how an agent understands your project's rules (the context)

Protocol. Framework. Context. One foundation. A coherent stack for building agents, not three random donations.

The catch nobody mentions

Linux Foundation projects are famously deliberate. "Deliberate" is the polite word for slow. MCP evolves at startup speed right now — weekly releases, fast iteration. If AAIF governance turns that into quarterly committee meetings, the community will fork it — copy the code and evolve it independently, leaving the "official" version behind.

Ninety-seven companies joining in under three months proves the demand is real. Enterprises want standards like oxygen — without them, every AI agent integration is expensive custom work. With them, plug and play.

But demand without speed is just bureaucracy with a mailing list. The adults showed up. Now they need to keep up — because developers won't wait.