Google Cloud launches Gemini agent for work, now in private preview for enterprise customers

Google Cloud on Thursday, Oct. 8, introduced the Gemini agent, a single AI agent built to take work off an employee's plate and hand back something finished. CEO Thomas Kurian showed it during his keynote at Gemini at Work 2026, Google's annual enterprise AI event.

It is not generally available. The Gemini agent is in private preview for enterprise customers, and Google says wide availability for Workspace customers on select Business and Enterprise plans is coming soon. No date has been given.

What Google says it does

Google describes the product as a "universal agent for work." Users give it objectives rather than step-by-step instructions. It plans the job, uses tools, connects to a company's business systems and returns the result inside documents, the inbox and developer environments. It can answer questions, create content and media, and write and run code.

It runs in the cloud, so one set of memory and context follows a user across devices. "You never have to re-brief it or wonder which machine you told it to do something on," Kurian wrote in a company blog post.

Users can reach it from the Gemini Enterprise app, from inside Gmail, Docs, Drive, Sheets, Slides, Chat and Calendar, and from third-party services including Slack and Microsoft 365. In Workspace apps, users can mention @Gemini to call it up. An API lets developers embed it in other applications without a dedicated interface.

Sub-agents and coworker agents

For bigger jobs, Gemini can spin up what Google calls "temporary, job-specific agents, each with their own identity." These sub-agents can coordinate parallel and sequential steps that run for hours or days.

Companies can also create "coworker agents." These act as team members with their own email address, in the format "@agents.company.com," and have access only to the information they are given. Kurian wrote that Gemini can serve as a personal assistant, or work on behalf of a group, "like a project manager within a team," or a specific role, "like an analyst in your finance department."

Claude under the hood

Google says the agent picks the best model for each task. Today that pool includes Google's own Gemini models and Anthropic's Claude models, with other private and open models to be added later. Google says the approach is meant to improve quality and lower costs.

Google also says the product has built-in cost controls and the security, administration and governance enterprise customers require. Those are the company's claims about its own product. No outside testing of them has been published.

Industry versions

Versions built for financial services and legal work are in preview now. Versions for government, healthcare and retail are listed as coming soon.

Those are the sectors where an agent with its own email address and access to internal systems carries the most risk, since the data involved is regulated and the cost of a wrong action is highest. Google has not detailed how much autonomy agents will have in those settings, or what human sign-off is required.

A crowded race

The launch follows rival releases within the past several weeks. OpenAI introduced "dots" in late September, always-on agents that pursue user goals across apps on their own. They are available to ChatGPT users on Pro and Business Premium accounts. Meta released Muse in September, a personal agent that can shop, book travel, send emails and make payments, and which CBS News reports has been downloaded millions of times.

Google already offers a consumer-facing agent, Gemini Spark, launched in May. It lives in a separate interface inside the Gemini app. The new enterprise agent is Google's attempt to fold that kind of capability into one work interface.

Alphabet shares were up marginally in early trading Thursday, according to Reuters.

What comes next

The open items are concrete. Google has to move the Gemini agent from private preview to general availability for Workspace customers on Business and Enterprise plans, and it has promised government, healthcare and retail versions. Customers will then find out what it costs and how it performs once it is running inside their own systems.