Google's Agent2Agent protocol has moved from pilot curiosity to production infrastructure faster than most cross-vendor standards do. At Google Cloud Next 2026, the company disclosed that A2A has reached 150 organizations running it in production, not testing — up from the roughly 50 partners at launch in April 2025. Microsoft, AWS, Salesforce, SAP, and ServiceNow are among the vendors reported to be running A2A in live environments, routing real tasks between agents built on entirely different platforms. The more interesting detail sits underneath the adoption number. The protocol is now governed by the Linux Foundation's Agentic AI Foundation and has reached version 1.2, with signed agent cards using cryptographic signatures for domain verification. That's a meaningful shift from A2A's earlier design. Agent Cards were originally just metadata documents describing an agent's capabilities and how to reach it, exchanged over plain HTTP and JSON. Making those cards signable means a receiving agent can now check not just what another agent claims to do, but whether the claim actually traces back to the domain it says it belongs to — closer to how TLS certificates let a browser verify a website's operator before trusting it. That distinction matters more as A2A becomes an interchange layer rather than a single-vendor feature. When agents from IBM, SAP, and a startup's custom stack are all handing off tasks to each other through the same protocol, the discovery and negotiation problem stops being about interoperable message formats and starts being about which of those agent cards you can actually believe. A2A already separates itself from Anthropic's Model Context Protocol by design — MCP handles agent-to-tool access, A2A handles agent-to-agent coordination — and the two are explicitly meant to be run together rather than compete. The GitHub Foundation stewardship also puts A2A on a governance track similar to other Linux Foundation infrastructure: multi-vendor maintainership, public spec versioning, and a public issue tracker rather than one company's roadmap. That's a real signal for any team evaluating whether to build against it long-term, since protocol churn under single-vendor control is one of the standard risks in this space. What's still unresolved is verification depth. A signed domain claim confirms an agent card originated from a given domain; it doesn't yet attest to what that agent has actually done with a task once accepted. The identity layer is filling in faster than the accountability layer around it — a gap other agent-identity efforts, including W3C's new Agent Identity Registry work, are trying to address from a different angle.