Google Rebrands Vertex AI as Gemini Enterprise Agent Platform, Puts A2A Protocol at Center of Multi-Agent Strategy

Something interesting happened at Google Cloud Next 2026 that did not get the announcement treatment it deserved. Vertex AI — the platform that tens of thousands of developers used to build LLM applications without needing to think too hard about infrastructure — is gone. In its place: the Gemini Enterprise Agent Platform. That is not a rebrand in the marketing sense. It is an architectural admission that the market moved, and Google is now reorganizing to catch up rather than lead.

The rebranding comes with a June 24, 2026 deadline that should be on every Vertex AI developer's calendar. Deprecated SDK modules stop receiving updates on that date. Existing services keep running — Google is careful to say that — but anyone building new agentic workflows on the old SDK is making a decision to be on an unpatched stack in an area that is evolving fast. That is its own kind of risk, even if it does not feel urgent today.

The more consequential story is the A2A protocol. Google developed the Agent-to-Agent protocol and donated it to the Linux Foundation early in 2026. The strategic logic was not altruistic. If the industry converges on a standard protocol for agents to talk to each other, Google positions itself as the foundation provider rather than just another framework vendor. The alternative — letting some other entity own the interop layer — would be to cede the referee role to a competitor.

The framework support table in the announcement tells you how serious this bet is. Google ADK, LangGraph (official connector), CrewAI (native), LlamaIndex Agents (native), Semantic Kernel (native), and AutoGen (native) all have A2A support as of April 2026. That is five major frameworks plus Google's own tooling agreeing on the same wire protocol. MCP connects agents to tools. A2A connects agents to agents. They are complementary, and together they are starting to look like the TCP/IP stack for AI agent infrastructure — not because anyone planned it that way, but because the industry needed it badly enough that it converged anyway.

Version 1.2 of A2A shipped in March 2026, and more than 150 organizations have it running in production as of this month. That is meaningful early traction for a protocol that has been public for only a few months. It suggests that the pain of building agent systems without a standard communication layer was acute enough that shops did not wait for the spec to be perfect before adopting it.

The eight platform components that Google is reorganizing around are worth understanding individually, because they represent an explicit theory of what enterprise agent infrastructure needs to look like. Agent Studio is the low-code visual canvas for building agents without writing code. Agent Runtime handles long-running stateful agents. Agent Memory Bank persists cross-session context with low-latency retrieval. Agent Gateway enforces security policy. Agent Identity gives each agent a cryptographic ID with audit logging. Agent Registry is an internal catalog. Agent Observability provides structured logging and distributed tracing. Agent Simulation lets teams inject failure scenarios and stress-test multi-agent coordination before production. None of these are revolutionary ideas. What is notable is that Google is building all of them instead of letting the ecosystem assemble them from separate vendors.

ADK v1.0 achieving stable release across Python, TypeScript, Go, and Java with identical feature parity at the protocol level is also worth pausing on. Multi-language support for agent frameworks usually means "we have SDKs in multiple languages but they behave differently." Google is claiming feature parity, which is the harder engineering promise. If the parity holds in practice, it matters for the same reason A2A matters: enterprises with polyglot stacks need to know that their agent infrastructure will behave consistently regardless of which language a given service is written in.

There is a counter-argument that deserves attention. Google has announced many developer platforms before that subsequently lost momentum. The Gemini Enterprise Agent Platform is real in the sense that it exists and has components you can use today. But the enterprise agent market is also still early enough that the winning architecture has not been proven out at scale. Google is reorganizing around a bet on what agent infrastructure looks like. That bet could be right, but it could also be wrong in ways that will not be apparent for another year or two.

The practical takeaway for builders is straightforward. If you are starting a new agent project on Google Cloud, the Gemini Enterprise Agent Platform with ADK is the right starting point. The migration deadline should push existing Vertex AI developers to evaluate the new SDK sooner rather than later, even if the old services still work. If you are building cross-framework agent systems, A2A support across LangGraph, CrewAI, and Microsoft Agent Framework is real and worth designing around. The protocol is stable enough to bet on, and the cost of waiting for it to be more mature is higher than the cost of adopting it early.

Google made a quieter but potentially more durable move than the rebranding suggests. By donating A2A to the Linux Foundation, they chose to compete on platform quality rather than protocol ownership. Whether that discipline holds as the market matures is the more interesting question — and the one that will determine whether the Gemini Enterprise Agent Platform becomes the infrastructure layer Google is betting it will be.

Sources: Dev.to — Google Gemini Enterprise Agent Platform, Google Cloud Documentation, Google ADK