Google cedes control of A2A, handing rivals a shared agent standard

Google no longer controls the rulebook for how AI agents talk to each other. It has handed that job to a neutral body that already hosts Anthropic's rival infrastructure protocol, a move that strips Google of a potential lock-in lever just as enterprises start building multi-vendor agent systems.

The Agentic AI Foundation (AAIF), part of the Linux Foundation, announced on 20 August that it had taken over Google's Agent2Agent (A2A) protocol. A2A's own blog confirmed on 27 August that the project had been accepted as a Growth Stage project within the foundation, according to the A2A Protocol blog.

A2A launched in 2025 as an open protocol letting AI agents from different vendors find each other, communicate and hand off tasks. More than 150 organisations already back it. It now sits alongside Anthropic's Model Context Protocol (MCP), goose and AGENTS.md inside AAIF, a foundation whose membership has grown from fewer than 40 at its December 2025 launch to more than 250 today, including AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI, according to Techzine.

The two protocols are complementary rather than competing: MCP connects an agent to tools and data, a vertical link, while A2A lets agents work with each other as peers, a horizontal one. Governance splits between a Governing Board handling strategy, budget and membership, and a Technical Committee running projects and reviews; the Linux Foundation supplies legal and operational structure but no technical direction. Google has likened the approach to Android: open standards paired with an ecosystem, designed to speed adoption rather than slow it behind one company's roadmap.

Mazin Gilbert, AAIF's executive director, said "no single company will build every agent an enterprise needs."

Placing A2A next to MCP under one governance structure removes the biggest practical objection enterprises had to building agent systems that span vendors: incompatible plumbing. That benefits every company that isn't Google or Anthropic, because it means a buyer can mix agents from AWS, Microsoft, OpenAI and others without betting the architecture on one supplier's protocol surviving.

For Google, the transfer trades control for adoption. Keeping A2A in-house would have let it shape the standard to its own products' advantage, but a proprietary protocol competing against MCP's growing footprint risked fragmenting the market and slowing uptake of A2A itself. Ceding governance to AAIF, where its own protocol now sits next to a foundation with 250-plus members, effectively trades ownership for the certainty that A2A becomes the standard rather than a standard. Anthropic, whose MCP was there first, does not have to fight A2A for dominance; it gets a peer protocol under the same neutral roof, which is a cheaper outcome than a standards war.