Google gives agents an identity, an inbox and a spend cap

What it does

Google Cloud announced the Gemini agent on Thursday, October 8, at its Gemini at Work 2026 event, in a post adapted from the keynote by Thomas Kurian, CEO of Google Cloud. The pitch is one agent instead of a fleet — “your new single, universal agent for work” — answering questions, handling knowledge work, creating images and media, and writing and running code from a single prompt box and a single API. “You give it objectives, not instructions.” Because execution happens in the cloud it keeps one set of memories, one context and a single personalisation graph across the web, iOS, Android, Windows, Mac, the command line, Google Workspace, Microsoft 365 and Slack; work that runs for hours or days keeps going after you close your laptop, and it can operate headless inside a third-party application with no dedicated interface at all. Two structural ideas carry it. The first is multi-agent orchestration : Gemini can dynamically create a roster of sub-agents — temporary, job-specific, each with its own identity — and coordinate parallel and sequential steps across them. The second is the coworker agent . You describe the role you need and Gemini creates it, and the agent receives its own Workspace account including an email address, a calendar, Drive storage and a presence in your company directory. Colleagues work with it the way they work with anyone else, adding it to a Chat space or @mentioning it; it can suggest an edit in a Doc and reply in the comment thread, appearing under its own name in version history. Google is explicit that the agent “acts under its own identity rather than yours, and it sees only what you share with it.” Underneath, the model is a choice rather than the product: it runs each job on the model that fits best, “orchestrating across our Gemini family of models and Claude models from Anthropic today, and other leading private and open models in the future.” The governance section is the longest in the post and is organised around four questions — who is the agent, what is it allowed to do, what did it do, and what should it never touch. Every agent gets its own cryptographically attested identity , governed like an employee with least-privilege permissions, stamped into the logs that capture its work and into any virtual machine spun up to run code on its behalf. Permissions are fine-grained and role-based, approved by your security administrators, and mapped through standards such as OAuth when the agent reaches an external system. Every action is written to an audit trail attributed to the agent rather than to a person . All agents execute inside an Agent Sandbox with its own network boundary, and all traffic — in, out and between agents — passes through Agent Gateway, an AI network firewall enforcing your policies in real time. You write a policy once, such as “agents may not open documents classified Need to Know,” and it applies to every agent in the company instead of being checked one agent at a time. On cost, the post pairs multi-model orchestration and automatic smart routing with real-time spend caps : you set a hard limit on a project’s AI spend in the Cloud Billing Console, Gemini monitors token usage and sandbox costs, and if the cap triggers, that project’s agent pauses until you choose to resume it with a single click. On availability the post itself states no stage and no price for the Gemini agent; Google confirmed to VentureBeat and The Verge that it is in private preview to a select group of customers as of October 8, with wider availability promised and no date given. The only item on the page carrying its own stated stage is industry specialisation, in preview for Financial Services and Legal.