Gemini agent runs your work from one prompt across apps

The Gemini agent is a single AI agent for work, aimed at business customers, that plans a task, uses tools and connected company systems, and returns finished results inside your documents, inbox or coding tools. Google’s pitch is that you give it objectives rather than step-by-step instructions.

It can answer questions, handle research and other knowledge work, create media, and write and run code. It lives in the Gemini Enterprise app, inside Workspace and in some third-party services. Google says it has the security, admin and governance controls big companies require, plus built-in limits on spending.

To call the agent into a conversation or file, you type @Gemini, which works across Chat, Gmail, Slides, Sheets and Docs. It also works from your phone, your desktop, the web, a command line, Slack and Microsoft 365. Because it runs in the cloud, it keeps the same context whichever device you use.

Google gave two everyday examples:

• The slide deck request: if your manager emails asking for a project update as a presentation, Workspace Intelligence spots it as a task you could hand off and offers a one-click way to pass it to Gemini.
• The meeting nobody wants to organize: ask Gemini to set up next week’s meeting with the usual regional event leads, without naming anyone. It works out who they are from your chat space and the thread from your last event, checks calendars and starts an email thread to find a time, even with people outside the company.

Developers can also plug the agent into other software through an API, running it as a headless agent, which means it can work without a screen of its own.

The most striking feature is the coworker agent. A business describes the role it wants filled, such as an events coordinator or a finance analyst, and Google creates an agent with its own Workspace account. That includes an email address, a calendar, Drive storage and a listing in the company directory.

In Google’s example, a marketing manager asks an events coordinator agent in Chat to prepare a launch-readiness document. A colleague then tags the agent in a Docs comment to request a change. Its replies and edits show up under its own name and in the document’s version history, just like a human teammate’s.

Gemini can also work as a team member on behalf of a group, like a project manager, rather than only as one person’s assistant.

The Gemini agent isn’t tied to Google’s own AI. It runs each job on whichever model fits best, and today that includes both Google’s Gemini models and Anthropic’s Claude models. Google says support for other private and open models will follow. The goal, it says, is better quality at a lower cost.

For big jobs, it splits the work among sub-agents, temporary helpers that each handle a piece, and coordinates them in parallel or in sequence. Those jobs can run for hours or days and keep going after you close your laptop.

To stay on track, it keeps four kinds of memory:

• Session memory for the current task, even one that runs for days.
• Semantic memory, a knowledge base built from documents, conversations and other agents.
• Procedural memory for how jobs get done, including skills it writes for itself.
• Episodic memory, a record of everything it has done before.

Administrators can set project spending caps that pause an agent when its cloud costs hit a preset limit.

If your company pays for Google Workspace on an eligible plan, you may soon be able to hand off the tedious middle part of tasks, like gathering names, checking calendars and drafting the first version. That’s useful, but it doesn’t remove the need to check the result.

• Check its work. Compare any figures in a report or update against the original documents before you share it.
• Know the rules. Expect your employer to decide which tasks an agent can finish alone. Customer emails and financial changes should still get human review.
• Watch what it can see. A coworker agent with directory access and its own inbox can reach a lot of company information, so IT teams should test permissions and audit logs before rollout.

The Gemini agent is in private preview for enterprise customers. Google says wide availability is coming soon for Workspace customers on select Business and Enterprise plans, but it hasn’t given a date. Some features, including long-running background tasks, mobile and desktop access, and choosing third-party models, are listed as early access. Businesses can try Gemini Enterprise Plus free for 30 days, though that doesn’t guarantee access to every new capability.