Google rolls out Gemini agent platform

Google Cloud used its Next 2026 conference to launch Gemini Enterprise Agent Platform, a new system for building, running, governing, and optimizing AI agents. (cloud.google.com) An AI agent is software that can take instructions, call tools, and complete multi-step work with limited human input. Google said companies are now asking less whether they can build one agent and more how to manage thousands. (blog.google) (cloud.google.com) Google said the new platform is the evolution of Vertex AI, its existing machine learning service. Future Vertex AI services and roadmap updates will be delivered through the new Agent Platform rather than as a standalone service. (cloud.google.com) The stack includes Agent Studio for low-code development, an upgraded Agent Development Kit for code-first teams, Agent Runtime for long-running agents, and Memory Bank for persistent context. Google also said customers can access more than 200 models through Model Garden, including Gemini, Gemma, and Anthropic’s Claude models. (cloud.google.com) Google framed the product as a control layer as much as a model layer. Sundar Pichai called it “mission control for the agentic enterprise,” with governance, security, and optimization built into the same platform. (blog.google) The company paired the launch with fresh usage numbers. Pichai said Google’s first-party models now process more than 16 billion tokens per minute through direct customer API use, up from 10 billion in the prior quarter. (blog.google) Google also said paid monthly active users of Gemini Enterprise rose 40% quarter over quarter in the first quarter. In its February earnings remarks, the company had said Gemini Enterprise had sold more than eight million paid seats four months after launch. (blog.google 1) (blog.google 2) At the conference, Google showed how the platform is supposed to work in practice with a marathon-planning demo in Las Vegas. The demo used a planner agent, an evaluator agent, and a simulator agent connected through Google’s development kit, runtime, registry, and protocols. (cloud.google.com) Bain said the launch shows Google is repositioning from selling model access to selling an enterprise control plane for agents. Its analysts highlighted governance features such as Agent Identity, Agent Registry, Agent Gateway, Agent Simulation, Agent Evaluation, and Agent Observability. (bain.com) That puts Google closer to the center of how companies deploy AI at work: not just picking a model, but deciding which agents can act, what data they can use, and how their behavior is monitored. (bain.com)