Google on Thursday, Oct. 8, unveiled the Gemini agent, an AI agent for businesses: AI that can carry out tasks on its own, such as using apps and websites, instead of only answering questions. Thomas Kurian, who runs Google Cloud, introduced it at the company's Gemini at Work conference. Google says workers give it a goal rather than step-by-step instructions and come back to finished work. From a single box, Google says, it can answer questions, take on writing and research work, make images and other media, and write and run computer code.
It works inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, and can also be used on phones, on Windows and Mac computers, and through Microsoft 365 and Slack. Because it runs on Google's servers, Google says work that takes hours or days keeps going after someone closes their laptop.
Teams can also set one up as a "coworker agent." Google says it gets its own Workspace account, with an email address, a calendar, file storage and a place in the company directory, and colleagues work with it by adding it to a chat or tagging it in a document. It acts under its own name, not the name of the person who made it, and sees only what people share with it. Google says every action the Gemini agent takes, whether working for one person or a team, is recorded under the agent's name rather than a person's. Companies can write a rule once, such as "agents may not open documents classified Need to Know," and Google says it then applies to every agent in the company.
The agent does not run on Google's AI alone. Google says it picks the AI model, the underlying system that produces its answers and work, that best fits each job, choosing today between its own Gemini models and Anthropic's Claude models, with more models, including free-to-use open ones, to come. Google says a tool it calls Smart Routing sends each job to the model that does it well at the lowest cost, so simple jobs go to cheaper ones. Companies can set a hard spending limit for each project, and when a project hits its limit, that project's agent pauses until someone resumes it with one click. Versions for financial firms and for legal teams are available now as early test versions, and versions for government, health care and retail are coming, Google says.
Google's announcement gives no price and no date for the agent's wide release. According to a news report, for now only companies Google invites can try it, with a wider release planned for companies on some of Google Workspace's Business and Enterprise plans. Google says nearly 90 percent of the Fortune 100, the biggest U.S. companies by revenue, already use Gemini Enterprise, its existing business AI service. The customer success stories Google gave alongside the launch describe that older service, not the new agent. How well it handles jobs that run for hours or days, with no one checking its work, has not yet been shown outside Google.