Orchestration
A2A Protocol Hits v1.0 — And Leaves Google's House
The Agent2Agent protocol, Google's open standard for cross-vendor agent communication, shipped its 1.0 release and moved its governance to the Agentic AI Foundation (AAIF), an industry body distinct from the company that created it.
The announcement frames this as a maturity milestone: the protocol community is marking the release of A2A v1.0 as the first stable, production-ready version of the open standard for communication between AI agents. That's a meaningful shift from a single-vendor blog-post spec to something with a steering committee spanning multiple companies — the protocol is now guided by a technical steering committee with representatives from eight major technology companies, rather than decisions sitting inside one firm's roadmap.
The practical problem A2A is trying to solve hasn't changed: agents built on different frameworks need a shared way to discover each other, hand off tasks, and exchange state without a central orchestrator dictating the interaction. As organizations build increasingly sophisticated multi-agent systems, interoperability has become the defining challenge, and A2A addresses that challenge by combining support for multiple protocol bindings, seamless version negotiation, and a common semantic model so agents can interoperate across systems with predictable behavior.
What's worth watching is less the version bump than the governance move. A protocol that started as a single company's announcement is now explicitly positioned as neutral infrastructure: the AAIF's open agentic stack role means A2A provides an open standard for how autonomous AI agents discover each other, delegate tasks, and collaborate across distinct frameworks and vendor boundaries — language that puts A2A alongside MCP as foundation-governed plumbing rather than vendor product.
This matters for the orchestration layer specifically because version negotiation and semantic consistency are the unglamorous parts of multi-agent systems that actually break in production. A task handoff between an agent built on one framework and one built on another fails less often because of missing features than because of mismatched assumptions about message shape, retry semantics, or what counts as task completion. A 1.0 spec with a published steering committee gives implementers something stable to build conformance tests against — which is different from, and arguably more consequential than, the raw adoption numbers vendors like to cite.
The open question is whether foundation governance actually changes behavior when the dominant implementer is still the company that wrote the original spec. A steering committee with eight members is not the same as independence, and the real test will be whether non-Google proposals get merged over Google's objections. That's not yet demonstrated — it's a structure, not a track record.