OpenAI, Anthropic, and Block Collaborating to Foster Cooperative AI Agents

OpenAI, Anthropic, and Block have joined forces to establish a new open-source entity—the Agentic AI Foundation—aimed at fostering standards for artificial intelligence agents.

The trio is also handing over ownership of several widely utilized agentic technologies to the foundation. This encompasses Anthropic’s Model Context Protocol (MCP), which facilitates agent connectivity and interaction; OpenAI’s Agents.md, enabling programs and websites to define rules for coding agents; and Goose, a framework for agent development created by Block. While these technologies were already accessible at no cost, the foundation will allow for broader contributions to their evolution.

“MCP is utilized by numerous companies, but there are several others [that don’t adopt it],” notes Nick Cooper, who heads the protocol efforts at OpenAI. He believes that by making MCP an open standard, developers and enterprises will be more inclined to adopt it and create systems that incorporate agentic AI. “That open interoperability—that open standard—essentially allows companies to communicate across providers and agentic systems.”

The Agentic AI Foundation will operate under the Linux Foundation, which manages the development of the popular open-source Linux operating system along with various other projects. The foundation will provide both legal and technological backing for the establishment of open-source initiatives. Additional companies supporting the AAIF, alongside the three founding organizations, include Google, Microsoft, AWS, Bloomberg, and Cloudflare.

This new foundation signifies an emerging transition from chat-oriented AI systems towards increased use of applications that perform actions on behalf of users. Such agentic AI holds the promise of a potentially profitable new framework where AI agents interact on the web and negotiate with each other to enable a variety of applications. For instance, consumers might enlist AI assistants to make purchases and bookings, while businesses could employ AI agents to handle transactions and customer engagements.

Srinivas Narayanan, OpenAI’s chief technology officer for B2B applications, imagines a future where numerous AI agents routinely communicate during business operations. The collaboration within the AI industry, under shared open standards, should facilitate smooth interactions. “Open source is poised to play a significant role in shaping and adapting AI in the real world,” Narayanan states.

The issue of openness is crucial to the current landscape of AI. Most US companies generate revenue by providing access to powerful closed models via application programming interfaces, or APIs. Meta had previously shared the weights for its top model, Llama, allowing anyone to download and run it, although the company has recently indicated a shift toward a more closed model. In contrast, several Chinese AI firms, such as DeepSeek, Alibaba, Moonshot AI, and Z.ai, offer strong open-source models that are gaining popularity among developers, startups, and AI researchers. Some express concerns that this scenario may provide Chinese companies with a significant strategic edge over time.