Google Gives Gemini a Work Identity and an Email Address

Google has introduced a universal Gemini agent for businesses, reportedly in private preview for enterprise customers. The pitch: instead of asking a chatbot a question, you hand a task to an agent, meaning an AI that carries out steps on its own, and it works on it in the cloud while you do something else.

What was announced

According to coverage that draws on The Verge, the agent lives in the Gemini Enterprise interface and keeps its context across your phone, desktop and browser sessions. It can use job-specific sub-agents, picks the model it judges best for each task, and connects to Google Workspace apps (Gmail, Drive, Docs, Sheets, Calendar) as well as outside tools such as Slack and Microsoft 365. The most eye-catching piece is the “coworker agent”: a configuration that gets its own identity and a company email address, so colleagues can send it work much like they would a person. No pricing was reported, and since it is a private preview, broad availability is unconfirmed.

What is behind this

Giving an agent its own identity solves a real problem. Today, most AI helpers act through your account, which makes it hard to see who did what, or to limit what they can touch. A separate identity lets a company set permissions, read logs and revoke access, the same way it would for a new hire. It also echoes a worry in the news this week: Wikimedia says agents it believes were OpenAI’s behaved unpredictably and were hard to identify. Clear identity is part of the answer to that, though not the whole answer.

A fair caveat is that most of the detail comes from secondary reporting, and Google has not shown how reliably multi-step jobs finish. Companies usually test connector permissions, audit logging and error handling for months before trusting agents with real work.

What this means for you: If you use Gemini personally, nothing changes today. If you work somewhere that runs Google Workspace, expect a pilot invitation sooner or later, and expect your IT team to have opinions. Good questions to ask: what can the agent read, who can see what it did, and what happens when it gets something wrong. For curious beginners, the useful mental model is a very fast junior colleague who never sleeps and needs clear instructions, not a magic button.

Sources

• Let’s Data Science: Google launches a universal Gemini agent
• AI Weekly: daily edition, October 8