OpenAI, Anthropic, and Block join the Linux Foundation’s new effort to unify the age of AI agents

As AI moves beyond chatbots and toward systems that can take action, the Linux Foundation is launching a new group dedicated to preventing AI agents from splintering into a mess of incompatible and closed products.

The group, called the Agentic AI Foundation (AAIF), will serve as a neutral home for open source projects related to AI agents. AAIF is supported at launch by donations from Anthropic, Block, and OpenAI.

Anthropic donates MCP (Model Context Protocol), a standard way to connect models and agents to tools and data; Block contributes to Goose, its open source proxy framework; OpenAI brings AGENTS.md to the table, and developers of its simple instruction files can add them to a repository to tell AI coders how to behave. You can think of these tools as the basic plumbing of the customer era.

Other members of the AAIF include AWS, Bloomberg, Cloudflare, and Google, signaling an industry-wide push for common guardrails so AI agents can be trustworthy at scale.

In the view of OpenAI engineer Nick Cooper, protocols are essentially a common language that allows different agents and systems to work together without each developer reinventing integrations from scratch.

“We need multiple [protocols] To negotiate, communicate and work together to deliver value to people, and this kind of openness and communication is why it will never be one provider, one host or one company.

Jim Zemlin, executive director of the Linux Foundation, put it more bluntly in talks about the launch: The goal is to avoid a future of “closed wall” proprietary stacks, where tool communication, agent behavior, and orchestration are locked behind a few platforms.

“By bringing these projects together under the AAIF, we are now able to harmonize interoperability, safety patterns, and best practices specifically for AI customers,” Zemlin said.

Block — the fintech company behind Square and Cash App — isn’t known for its AI infrastructure, but it’s playing open with Goose. AI technology lead Brad Axen paints it as proof that open alternatives can match proprietary proxies at scale, with thousands of engineers using them weekly for coding, data analysis and documentation.

The open source Goose serves a dual purpose for Block.

“Getting it out into the world gives us a place for other people to come and help us improve it,” Axen told TechCrunch. “We have a lot of open source contributors, and everything they do to improve it goes back to our company.”

Meanwhile, Goose’s donation to the Linux Foundation gives Block access to community stress testing while positioning it as a working example of AAIF’s vision — an agent framework designed to plug into common building blocks like MCP and AGENTS.md.

Anthropic is making a similar move at the protocol layer, handing over MCP to the Linux Foundation. The goal: to make MCP the neutral infrastructure that connects AI models to tools, data, and applications without endless one-time adapters.

“The main goal is to get enough adoption in the world to be the de facto standard,” David Soria Parra, co-founder of MCP, told TechCrunch. “We’re all better off if we have an open integration center where you can build something as a developer and use it across any client.”

The donation of MCP to the AAIF indicates that the protocol will not be controlled by a single vendor.

This governance point is key to why the Linux Foundation would create a new umbrella at all. The organization already hosts major AI and developer infrastructure projects — everything from PyTorch and Ray to Kubernetes — but says the AAIF specifically aims for agent and orchestration standards, including common safety patterns and interoperability.

The AAIF structure is funded through a “directed fund,” meaning companies can contribute money through membership fees. But the Linux Foundation’s Zemlin argues that funding does not equal control: project roadmaps are drawn up by technical steering committees, and no single member has a say in the direction.

However, the big question is whether the AAIF will become a true infrastructure or just another industry logo alliance.

“An early indicator of success, in addition to the adoption of these standards, is the development and implementation of common standards used by seller agents around the world,” Zemlin said.

For OpenAI’s Cooper, success might look like an evolution of standards: “I don’t want them to be stagnant. I don’t want these protocols to be part of this foundation, and that’s where they’ve stayed for two years. They have to constantly evolve and accept more input.”

There is also a more nuanced consequence: even with open governance, implementation by a single company can become the default simply because it ships faster or gets the most usage. This isn’t necessarily a bad thing, Zemlin says. He points to the history of open source – such as Kubernetes “winning” the container race – as evidence that “dominance comes from merit, not control of vendors.”

For developers and enterprises, the short-term appeal is clear: less time creating custom connectors, more predictable agent behavior across codebases, and simpler deployment in security-conscious environments.
The bigger vision is more ambitious: If tools like MCP, AGENTS.md, and Goose become standard infrastructure, the agent landscape could shift from closed platforms to an open, hybrid software world reminiscent of the interoperable systems that built the modern Internet.