As AI advances beyond chatbots to systems capable of executing tasks, the Linux Foundation is initiating a new organization focused on preventing AI agents from diverging into a chaotic array of incompatible, proprietary products.
This new organization, named the Agentic AI Foundation (AAIF), will serve as a neutral hub for open-source projects related to AI agents. Initial support for the AAIF comes from contributions by Anthropic, Block, and OpenAI.
Anthropic is contributing its MCP (Model Context Protocol), which offers a standardized method for linking models and agents to various tools and data sources. Block is providing Goose, its open-source agent framework, and OpenAI is bringing AGENTS.md, a simple instruction file for guiding AI coding tools. These tools can be considered the fundamental infrastructure for the emerging agent era.
Other participants in the AAIF include AWS, Bloomberg, Cloudflare, and Google, indicating a broad industry push to establish common safeguards that ensure the reliability of AI agents at scale.
According to OpenAI engineer Nick Cooper, protocols essentially function as a shared language, enabling different agents and systems to collaborate efficiently without requiring every developer to create custom integrations from scratch.
“We require multiple [protocols] to negotiate, communicate, and work together to deliver value for users, and this kind of openness and communication is why it will never be dominated by a single provider, host, or company,” Cooper explained to TechCrunch.
Jim Zemlin, the executive director of the Linux Foundation, stated more directly during discussions about the launch that the objective is to avert a future characterized by “closed wall” proprietary ecosystems, where tool connections, agent behaviors, and orchestration are restricted to a select few platforms.
“By consolidating these projects under the AAIF, we can now coordinate interoperability, safety patterns, and best practices specifically for AI agents,” Zemlin remarked.
Block—the fintech company behind Square and Cash App—is not primarily known for AI infrastructure, yet it is making a commitment to openness with Goose. Brad Axen, AI Tech Lead, views it as proof that open alternatives can compete effectively with proprietary agents at scale, evidenced by thousands of engineers using it weekly for coding, data analysis, and documentation.
Open-sourcing Goose serves a dual purpose for Block.
“Making it available to the world provides an avenue for others to help us enhance it,” Axen told TechCrunch. “We benefit from numerous open-source contributors, and all their improvements come back to our company.”
Concurrently, donating Goose to the Linux Foundation provides Block with access to community stress-testing while positioning it as a tangible example of AAIF’s vision—an agent framework designed for seamless integration with shared components like MCP and AGENTS.md.
Anthropic is adopting a similar strategy at the protocol layer, entrusting MCP to the Linux Foundation. The objective is for MCP to become the neutral standard infrastructure for connecting AI models to tools, data, and applications, thereby eliminating the need for countless custom adapters.
“The primary goal is for it to achieve sufficient global adoption to become the de facto standard,” MCP co-creator David Soria Parra told TechCrunch. “We all benefit from having an open integration hub where developers can build something once and utilize it across any client.”
The donation of MCP to AAIF signifies that the protocol will not be controlled by a single vendor.
This emphasis on governance is a core reason why the Linux Foundation established an entirely new umbrella organization. While the foundation already hosts major AI and developer infrastructure projects—ranging from PyTorch and Ray to Kubernetes—AAIF is specifically tailored for agent standards and orchestration, encompassing shared safety frameworks and interoperability.
AAIF’s operational model is financed via a “directed fund,” allowing companies to contribute financially through membership fees. However, Zemlin of the Linux Foundation asserts that funding does not equate to control: project roadmaps are determined by technical steering committees, and no single member possesses unilateral authority over the direction.
Nevertheless, the critical question remains whether AAIF will evolve into genuine infrastructure or merely another symbolic industry alliance.
“A key early indicator of success, beyond the adoption of these standards, would be the development and deployment of shared standards by vendor agents worldwide,” Zemlin noted.
For OpenAI’s Cooper, success would involve the continuous evolution of standards: “I don’t want it to be static. I don’t want these protocols to sit within this foundation unchanged for two years. They should develop and consistently incorporate new input.”
There is also a more subtle implication: even with open governance, one company’s implementation might become the default simply by being the fastest to market or gaining the most widespread use. Zemlin, however, considers this not inherently negative. He references open-source history—like Kubernetes’s triumph in the container ecosystem—as evidence that “preeminence arises from merit, not vendor control.”
For developers and enterprises, the immediate advantages are clear: reduced time spent on building custom connectors, more predictable agent behavior across various codebases, and simpler deployment within security-conscious environments.
The broader vision is more ambitious: if tools such as MCP, AGENTS.md, and Goose become widely adopted infrastructure, the agent landscape could transform from closed platforms into an open, modular software environment, reminiscent of the interoperable systems that forged the modern web.