As AI evolves from basic chatbots to systems capable of performing actions, the Linux Foundation is launching a new organization aimed at preventing AI agents from fragmenting into a collection of incompatible, proprietary products.
Dubbed the Agentic AI Foundation (AAIF), this initiative will serve as a neutral home for open-source projects related to AI agents. Its launch is backed by contributions from Anthropic, Block, and OpenAI.
Anthropic is contributing its Model Context Protocol (MCP)—a standardized method for connecting models and agents to tools and data. Block is donating its open-source agent framework, Goose. Meanwhile, OpenAI is contributing AGENTS.md, a lightweight specification file that developers can include in their repositories to guide AI coding tools on how to interact with their codebases. Together, these components form foundational plumbing for the emerging agent era.
Additional AAIF members include AWS, Bloomberg, Cloudflare, and Google—highlighting broad industry alignment around shared safeguards to ensure AI agents can be trusted at scale.
According to Nick Cooper, an engineer at OpenAI, protocols function as a common language that enables diverse agents and systems to collaborate without requiring every developer to reinvent integration from scratch.
“We need multiple [protocols] to negotiate, communicate, and cooperate to deliver real value to people. This openness and interoperability are precisely why there will never be just one provider, one host, or one company,” Cooper told TechCrunch.
Jim Zemlin, Executive Director of the Linux Foundation, was even more direct in discussions about the launch: the goal is to avoid a future dominated by “walled garden” proprietary stacks where tool integrations, agent behaviors, and orchestration are locked behind a few dominant platforms.
“By bringing these projects under the AAIF umbrella, we can now coordinate interoperability, security patterns, and best practices specifically tailored for AI agents,” Zemlin said.
Block—the fintech firm behind Square and Cash App—is not typically associated with AI infrastructure, but it has made a significant push toward openness through Goose. Brad Axen, Block’s head of AI, sees it as proof that an open alternative can rival proprietary agents at scale, with thousands of engineers using it weekly for coding, data analysis, and documentation.
Open-sourcing Goose serves a dual purpose for Block.
“Releasing it publicly gives others a platform to help us improve it,” Axen told TechCrunch. “We’ve received numerous contributions from the open-source community, and all those enhancements flow directly back to our internal workflows.”
Donating Goose to the Linux Foundation also subjects it to community-driven stress testing while positioning it as a working example of AAIF’s vision—an agent framework designed to plug into shared building blocks like MCP and AGENTS.md.
Anthropic is taking a similar approach at the protocol level by handing over MCP to the Linux Foundation. The aim is to establish MCP as vendor-neutral infrastructure for linking AI models to tools, data, and applications—replacing the current patchwork of one-off adapters.
“The primary goal is to achieve enough global adoption for it to become the de facto standard,” MCP co-creator David Soria Parra told TechCrunch. “If we have an open integration hub where developers build something once and use it anywhere, everyone benefits.”
Contributing MCP to AAIF signals that the protocol won’t be controlled by any single vendor.
This governance model lies at the heart of why the Linux Foundation created this new umbrella organization. While it already hosts major AI and developer infrastructure projects—from PyTorch and Ray to Kubernetes—it says AAIF is uniquely focused on agent-specific standards and orchestration, including shared security models and interoperability frameworks.
AAIF is funded through a “directed fund” structure, allowing companies to contribute via membership fees. However, Zemlin emphasizes that funding does not equate to control: project roadmaps are set by a technical steering committee, and no single member can unilaterally dictate direction.
The big question remains: will AAIF become genuine infrastructure—or just another industry consortium with symbolic backing?
“Early indicators of success, beyond adoption of these standards, include the development and implementation of shared specifications used by agent platforms worldwide,” Zemlin said.
For OpenAI’s Cooper, success means continuous evolution: “I don’t want this to become stagnant. I don’t want these protocols to join the foundation and then sit untouched for two years. They must keep evolving and incorporating broader input.”
There’s also a subtler dynamic at play: even under open governance, one company’s implementation could become the de facto default simply by shipping first or gaining the most traction. Zemlin argues this isn’t inherently negative, pointing to open-source history—like Kubernetes dominating container orchestration—as evidence that “leadership emerges from merit, not vendor control.”
For developers and enterprises, the near-term benefits are clear: less time spent building custom connectors, more predictable agent behavior across codebases, and simpler deployment in security-conscious environments.
