Google demos Gemini agent platform

Google Cloud used its Next ’26 conference in Las Vegas on April 22 to launch Gemini Enterprise Agent Platform, a new system for building, running, governing and monitoring AI agents. (cloud.google.com 1) (cloud.google.com 2) Google describes the product as the evolution of Vertex AI, not a separate add-on. It said future Vertex AI services and roadmap updates will be delivered through Agent Platform instead of as a standalone service. (cloud.google.com) (virtualizationreview.com) The platform is organized around four jobs: build, scale, govern and optimize. Google’s launch materials highlight new pieces including Agent Studio, an upgraded Agent Development Kit, Agent Runtime, Agent Identity, Agent Registry, Agent Gateway, Agent Simulation, Agent Evaluation and Agent Observability. (cloud.google.com 1) (cloud.google.com 2) In plain terms, Google is trying to sell companies the plumbing for AI workers, not just the brains. The pitch is that an enterprise agent needs the same basics as any other production system: identity, policy controls, logs, testing, runtime management and access to company data. (cloud.google.com) (blog.google) That framing showed up in Google’s own language. Sundar Pichai wrote that the question has shifted from whether companies can build an agent to how they can manage thousands of them, while Thomas Kurian said Next ’26 was about moving AI into production across the enterprise. (blog.google) (cloud.google.com) Google tied the platform to scale numbers meant to show demand is already there. The company said nearly 75% of Google Cloud customers use its AI products, 330 customers processed more than 1 trillion tokens over the past 12 months, and 35 reached the 10-trillion-token mark. (cloud.google.com) (blog.google) The developer keynote turned that strategy into a concrete demo. Google’s Richard Seroter and Emma Twersky ran a live marathon-planning simulation for Las Vegas that used multiple agents to plan a route, model weather, traffic and crowds, and then run the race autonomously. (youtube.com) (github.com) Google also published that demo as open-source code in a GitHub repository called Race Condition. The repo says it was originally shown in the Next ’26 developer keynote and is meant as a reference architecture for autonomous agents built with Gemini and Google’s Agent Development Kit. (github.com) The repo adds a detail that matters for enterprise demos: reliability. Google says the project supports replaying recorded agent runs and includes a deterministic runner that mimics decision patterns without making live model calls, which lets teams test or present the system without turning every run into a billable experiment. (github.com) Google paired the platform launch with broader infrastructure announcements, including eighth-generation Tensor Processing Units, Agentic Data Cloud and new security products with Wiz. The package makes the agent platform look less like a chatbot feature and more like the control layer in a larger Google Cloud stack. (cloud.google.com) (virtualizationreview.com) By the end of Next ’26, Google said more than 32,000 people had attended the Las Vegas event. The headline from the week was that Google wants enterprises to treat agents as long-running workloads with names, permissions, memory and audit trails, not as one-off prompts. (cloud.google.com 1) (cloud.google.com 2)