OpenAI、Anthropic 与 Block 共同成立「Agentic AI Foundation」，并在 Linux 基金会旗下运作，将 Anthropic 的 MCP 协定、OpenAI 的 Agents.md 与 Block 的 Goose 代理框架捐出，以开放原始码方式推动代理型 AI 的共同标准。Google、Microsoft、AWS、Bloomberg、Cloudflare 等公司也加入成为成员。

新基金会希望建立跨公司、跨系统互通的开放标准，支援能替使用者自动执行任务的代理型 AI，而不只是聊天机器人。这将让企业与消费者能用 AI 代理处理订购、预订、交易与客服等操作，OpenAI 高层认为开源与开放标准将是 AI 在真实世界被广泛采用的关键。

文章也点出开放与封闭之间的角力：美国多数公司以封闭 API 模型赚钱，Meta 曾开放 Llama 权重后又转趋保守，同时中国的 DeepSeek、阿里巴巴、Moonshot AI 与 Z.ai 等则提供强力开源模型。若 AAIF 所推的开放代理标准成为全球主流，虽然技术上中立，仍可能让背后的美国公司在 AI 生态中取得长期影响力。

OpenAI, Anthropic, and Block have created the Agentic AI Foundation under the Linux Foundation, donating Anthropic’s Model Context Protocol, OpenAI’s Agents.md, and Block’s Goose framework to establish open-source standards for AI agents; other backers include Google, Microsoft, AWS, Bloomberg, and Cloudflare.

The foundation aims to define interoperable standards so agentic AI systems from different providers can talk to each other and act on users’ behalf for tasks like booking, purchasing, and customer service, reflecting a shift from chatbots toward action-taking software; OpenAI leaders argue that open source and shared standards will be crucial for real-world AI adoption.

The article also highlights tensions between open and closed approaches: most US firms monetize closed API models, Meta briefly opened Llama’s weights then moved toward more control, while Chinese companies like DeepSeek, Alibaba, Moonshot AI, and Z.ai offer strong open models that may confer strategic advantages; if AAIF’s tools and standards become dominant, they could, like web standards bodies, give the US companies behind them significant long-term influence over how AI is used worldwide.