At Google Cloud Next ’26 in Las Vegas, Google launched the Gemini Enterprise Agent Platform, a single product for building, scaling, governing, and optimizing AI agents. Most coverage treated it as a rebrand of Vertex AI with agent features bolted on. The most consequential line sits lower in Google’s own announcement: all Vertex AI services and roadmap evolutions will now be delivered “exclusively through the Agent Platform, rather than as a standalone service.” That is not a feature release. That is Google moving every enterprise AI customer it has onto a new control plane.
Reuters reported that Google is putting agents at the heart of its enterprise monetization push, and Google Cloud CEO Thomas Kurian was blunt about the phase change: “The experimental phase is behind us, and now the real challenge begins.” He also said Vertex AI usage recently shifted from old-style machine learning to “a sudden explosion in users building their own custom AI agents.”
What’s actually new
The platform is organized around four verbs. Build: a new low-code Agent Studio alongside the upgraded, code-first Agent Development Kit (ADK). Scale: a re-engineered Agent Runtime that supports long-running agents holding state for days at a time, backed by a persistent Memory Bank. Govern: Agent Identity, Agent Registry, and Agent Gateway, so every agent (including agents sourced from partners) gets a trackable identity and operates inside enterprise guardrails. Optimize: Agent Simulation, Agent Evaluation, and Agent Observability, with full execution traces into agent reasoning.
The model layer stays deliberately open. Model Garden exposes more than 200 models, including Google’s newest first-party releases (Gemini 3.1 Pro, Gemini 3.1 Flash Image, Lyria 3) and open-weight Gemma 4, plus third-party support for Anthropic’s Claude Opus, Sonnet, and Haiku. CNBC called it a one-stop shop for agents and chips, which is exactly how Google wants it read.
Governance is the battleground
Google’s framing is telling: agents are now “interacting across multiple systems, and often without security and governance guardrails.” That matches what security vendors have been shouting about for months. Autonomous agents are nonhuman actors with credentials, permissions, and no consistent identity layer, a gap we covered in Orchid Security’s agent identity research. Google’s answer is to make identity, registry, and gateway controls a native platform feature, so the fix for agent sprawl lives inside Google Cloud.
The competitive logic is straightforward. Per Reuters, OpenAI and Anthropic have been moving downstream from models into enterprise applications and agent-building tools. Google’s counter is not a better model argument, it is a full-stack argument: data, infrastructure, governance, and distribution, while happily hosting rivals’ models. Michael Gerstenhaber, the Google Cloud AI product VP hired from Anthropic, put it plainly: “the model companies will build models that we will partner with them to distribute, and we will help enterprises access the intelligence of those models.”
The signal: the control plane is the product
Look at where Google is open and where it is not. Openness lives at the model layer, where switching is already cheap and getting cheaper. Ownership lives at the operational layer: once your agent identities, registry entries, gateway policies, evaluation harnesses, and long-term agent memory all reside in Google’s control plane, changing vendors is no longer a model swap, it is re-platforming an entire fleet. That is the same gravitational pattern we mapped in how agentic AI creates lock-in, and it is exactly the dependency that procurement teams should be pricing in, per our agentic AI buyer’s guide.
The capex context makes the intent obvious. Sundar Pichai used the event to reaffirm a $175 billion to $185 billion capital spending plan for the year, with just over half going to cloud and machine-learning compute, and new TPUs pitched at roughly 80 percent better inference performance than the prior generation. Agent Platform is the layer where that spending is supposed to turn into durable, hard-to-cancel enterprise revenue.
The caveats matter. “Exclusively through the Agent Platform” means existing Vertex AI customers are being migrated whether they asked for it or not, and Google has published little pricing detail. The governance stack is announced, not proven: days-long stateful agents in production is a claim without a public track record yet. The launch-day customer quotes (Burns & McDonnell, Color Health) are endorsements, not case studies with numbers. And there is an unexamined conflict in the pitch itself: the vendor selling you agent governance is the same vendor whose platform consumption those agents drive. Your auditor is also your landlord.
What to watch
Three things. First, whether Google Cloud’s earnings commentary starts quantifying Agent Platform adoption rather than gesturing at it. Second, whether OpenAI and Anthropic answer with governance and identity layers of their own, which would confirm that the enterprise fight has moved from model quality to fleet management. Third, whether enterprises accept their cloud vendor as the identity authority for autonomous agents, or push for neutral, cross-platform standards. If a neutral standard fails to emerge in the next year, the control planes will have already hardened, and the switching costs will be baked in.
AI and business tech news, verified by a physician who reads the filings. One email a week, no noise.
