AINEWS
Google's A2A turns one. Now comes the hard part.
A2A, Google's protocol for letting AI agents hand off work to each other, marks its first birthday as a contested standard in a market that has not yet picked a winner.
A2A, Google's protocol for letting AI agents hand off work to each other, marks its first birthday as a contested standard in a market that has not yet picked a winner.
One year after Google launched A2A as the backbone of a "collaborative agentic ecosystem," the protocol landscape for AI agents talking to other AI agents is still genuinely unsettled, and a birthday blog post is not evidence that the enterprise has followed. A2A, which stands for Agent-to-Agent, is a software standard that lets autonomous AI systems discover each other, share context, and pass tasks across organizational boundaries without a human in the loop. Google published a one-year retrospective on Wednesday that frames the past twelve months as proof the concept is working, but the post is a vendor explainer, and the more useful story is what it leaves out: a competitive map, independent adoption data, and a clear answer to the question every buyer will eventually ask.
The pitch, in Google's telling, is that traditional APIs were built for rigid, deterministic software and cannot carry the "fluid and autonomous" behavior of modern agents. A2A is supposed to fill that gap with a common language for handoffs, identity, and context. The blog post, written by Google product, marketing, and customer-engineering staff working on the protocol, claims the "collaborative agentic ecosystem … is becoming a reality — and it's growing faster than you think" (How A2A is Building a World of Collaborative Agents). Both phrases are rhetorical, and a reader evaluating the protocol should treat them as marketing language until something other than the vendor confirms them.
The most important missing name in the post is Anthropic's MCP, or Model Context Protocol, which is the most visible competing standard in this space. MCP started as a way for a single model to pull in tools and data sources, and it has since been extended toward the same kind of multi-agent handoffs A2A targets. The two protocols are not identical in scope, and several large model and tooling vendors have publicly backed MCP, but Google's retrospective does not name it, weigh it, or explain how an enterprise should think about choosing between them. That silence is itself the story. A standard that is winning does not usually need to pretend the field is uncontested.
A second gap is adoption. Google characterizes the ecosystem as expanding fast, but the post offers no usage numbers, no independent deployment case studies, and no third-party measurements. The bylined authors are a senior technical product manager, a product marketing lead for AI agents, and a customer engineer, all of whom work on A2A at Google, which is fine for describing what the protocol is and what Google intends, and not fine for declaring that enterprises are running on it. A buyer who reads "growing faster than you think" and treats it as a market signal is being asked to accept Google's framing on Google's evidence.
A third gap is security. The post leans on the idea of a "black box" handoff, where one agent passes a task to another and the receiving agent is trusted to handle the context. That framing is closer to an enterprise sales pitch than to a verified property. In practice, agent handoffs create exactly the kind of trust boundary where prompt injection, context leakage, and unclear error handling can compound, and the post does not engage with any of it. If A2A's real architectural advantage is a clean handoff model, the proof would be a threat model, not a metaphor.
What to watch next is straightforward. First, whether MCP and A2A converge, fork, or settle into different layers of the same stack, since "agent protocol" is a category, not a single product. Second, whether any neutral body, a standards consortium, a major cloud provider, or a credible analyst firm, publishes adoption data that is not vendor-supplied. Third, whether the first widely reported enterprise incident tied to an agent handoff comes from an A2A deployment, an MCP deployment, or something else, because failure modes tend to clarify contested standards faster than press releases do.
The honest read of A2A's first year is that Google shipped a real protocol, real partners joined, and the engineering work deserves respect. It is also that "growing faster than you think" is a sentence Google wrote, not a measurement anyone else has confirmed, and that the next twelve months will be decided by the answers to the questions the birthday post chose not to ask.