Google Cloud is collapsing its enterprise AI pitch into a single product. At its Gemini at Work 2026 event on Thursday, the company introduced the Gemini agent — one agent for work that can answer questions, handle tasks, create content, and write code — in a launch reported by Reuters and laid out by Google Cloud CEO Thomas Kurian in a keynote and blog post.
The timing is not subtle. Tech companies are racing to sell AI that works autonomously rather than waiting for prompts, and Google’s rivals have moved first: OpenAI launched always-on agents in September that pursue a user’s goals across apps on their own, and Meta released a personal agent last month that can shop, book travel, send email, and make payments. Google’s answer is to argue that a company should not need a different agent for each of those jobs.
Objectives, not instructions
The pitch, as Kurian framed it, is that users should hand the agent objectives rather than step-by-step instructions. According to Google, the agent plans the work itself, uses tools, connects to a company’s business systems, and returns finished work inside documents, email, and developer environments. It runs persistently in Google’s cloud, so a job that takes hours or days keeps going after the user closes a laptop, and it can spin up temporary sub-agents, each with its own identity, to carry pieces of a larger task.
Distribution is the other half of the strategy. The agent works directly inside the Google Workspace apps people already live in, including Gmail, Docs, Sheets, and Calendar, and Google is also making it reachable through Microsoft 365 and Slack, across web, desktop, mobile, and the command line. An agent that sits inside a competitor’s productivity suite is a notable concession to how enterprises actually work — and a sign Google would rather own the agent layer than defend the app layer.
A coworker with an email address
The most striking piece of the announcement is what Google calls coworker agents: persistent agents that act as team members rather than tools. A coworker agent gets its own Workspace account, its own company email address, calendar, and Drive storage, and a seat in the company directory, so colleagues can delegate to it much as they would to a human teammate. Google stresses the boundary that comes with that identity: a coworker agent acts under its own identity rather than its creator’s, and it can see only the information that has been shared with it.
That design choice cuts both ways. Giving an agent its own identity makes its actions auditable and its access revocable — but it also means companies will be managing directory entries, permissions, and offboarding for software, a job most IT departments are only beginning to think about.
It will happily run a rival’s model
Under the hood, the agent does not commit to a single model. Google said it picks the model that fits each task, and today that means routing work across Google’s own Gemini family and — strikingly — Anthropic’s Claude models, with other models to be added later. For Google, whose model business competes directly with Anthropic’s, the message is deliberate: the value, and the lock-in, is meant to sit in the agent platform, the integrations, and the governance layer, not in any one model.
Governance is the sales pitch
That governance layer got as much attention as the agent itself, as the Futurum Group’s breakdown of the launch detailed. Agents carry their own identities, permissions propagate through OAuth, actions are recorded in per-agent audit trails, and agent traffic runs through sandboxing and gateway controls. Cost controls matter too: routing is designed to send work to a model that fits rather than the most expensive one available, and administrators can set real-time spend caps that pause an agent’s work when a project’s budget is hit.
Google is also starting with regulated industries. Versions of the agent built for financial services and legal work are now in preview, the company said, with versions for government, healthcare, and retail to follow.
What Google has not said
For all the completeness of the vision, the commercial details are still thin. The agent is in private preview, and reporting by 9to5Google indicates it will become widely available soon to Workspace customers on select Business and Enterprise plans. Google has not announced pricing or a general-availability date.
That leaves the launch where most enterprise agent launches now sit: the architecture is credible, the governance story is aimed squarely at the objections CIOs raise, and the real test — whether employees hand meaningful work to an agent with its own inbox, and whether the finished work holds up — starts only once customers can actually use it at scale. Alphabet shares, for their part, were up only marginally in early trading Thursday, suggesting investors are waiting for that same evidence.
