Google's Gemini Agent Gets an Email Address and a Job

For a few years now, talking to an AI has meant typing a question and reading an answer. On Thursday, Google announced a different arrangement. According to TechCrunch's report from Google Cloud's event, the company is launching a unified Gemini agent that doesn't just reply. It takes ownership of a task and gets it done. It even has its own email address, like a new hire.

The first rollout is for businesses, with consumers coming later. Below I walk through what Google says the agent can do and why it starts at work. I also look at the questions a coworker-shaped AI raises, including a few that Google's announcement doesn't answer.

From chatbot to colleague: what was announced

The core idea is a single agent that lives inside Gemini Enterprise and can plan and carry out work across a company's tools. TechCrunch describes it as a unified agent that can answer questions and also act on a user's behalf, all from one interface.

Thomas Kurian, CEO of Google Cloud, framed the shift this way: you can give it "objectives, not just instructions." That is a small phrase with a big implication. Instead of telling a tool exactly which button to press, you describe the outcome you want. The agent works out the steps, picks the skills and tools it needs, and connects to internal systems to get there.

Google isn't alone in this direction. TechCrunch notes that AI tools have been moving beyond conversation toward taking ownership of tasks like generating code, scheduling meetings and booking travel. It also points to consumer agents such as Meta's Muse and the recent debut of ChatGPT's Dots. Google's bet is that it can bring the same idea into the workplace at serious scale.

Why businesses go first

Many AI launches start with consumers and reach companies later. Google is doing the reverse. Sundar Pichai said at the start of the event that Gemini has more than 1 billion monthly active users. He also said nearly 90% of Fortune 100 companies already use Gemini Enterprise at work.

That installed base is a real advantage. An agent is only useful if it can reach where the work lives, and Google already has a foothold in many large organizations. Pichai also gave a more practical reason for starting in the enterprise. It lets Google tackle the "harder problems around security, scale, and performance" that come with powerful agents.

I find that candid, and a little reassuring. An agent that can act is riskier than a chatbot that can only talk. If a chatbot gets something wrong, you read a bad answer. If an agent gets something wrong, an email may already have been sent. Companies have IT departments, access controls and compliance teams. Consumers mostly have themselves. Learning where agents break in a controlled setting before releasing them to everyone is a sensible order of operations.

How the agent actually works

Here are the details TechCrunch reports. They show how much of this is plumbing rather than magic.

• Planning and delegation. The agent plans the work and can hand parts of it to subagents. A subagent is a helper focused on one slice of the job.
• Model choice. By default, the agent picks the best model for each task. Users can override that and choose one themselves, including third-party models, starting with Anthropic's Claude. Google says the picker will later expand to open-source and private models.
• Rich requests. A request can include attachments such as files, folders or projects built for specific workstreams, which combine files and skills.
• Broad connections. The agent can link to Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, Snowflake and others. It can also work securely with any Model Context Protocol (MCP) server, inside or outside the company network.

If MCP is unfamiliar, think of it as a common plug standard. It lets an AI connect to different tools and data sources without a custom adapter for each one. Support for it, plus the willingness to include a rival's models, suggests Google knows that most companies don't live in a single vendor's world.

A tasks inbox and a transparent mind

One feature I like is the "tasks inbox." Users can follow what Gemini is doing there. They can see its thinking process, how it delegates to subagents, which special skills it loads, the code it runs and its progress.

This matters because delegation only works when you can check up on the delegate. Anyone who has handed a project to a colleague knows the mix of relief and anxiety. You want it done, but you also want to know it's on track. An inbox that shows the work, rather than just the result, speaks to that anxiety directly.

There's a caveat, which is my own opinion. Seeing a model's reasoning is useful, but it isn't the same as understanding it. The visible steps are a good starting point for review. They don't guarantee that every step was right. A person still has to look.

The agent with its own Workspace account

The detail that grabbed the headlines is that the agent has its own Workspace account, as if it were another coworker. According to Google, as reported by TechCrunch, that means its own email address and its own context. It knows who is on which team, their time zones, who needs to approve what and what's on people's calendars.

You can call on it the way you would a person: tag it, email it, share a document with it or add it to a group chat. When it acts, it writes an audit trail attributed to the agent, not to a human.

I think this is the most interesting design choice in the announcement, and a thoughtful one. Giving the agent its own identity solves a real problem. If an AI acts through your login, everything it does looks like you did it. A separate identity makes its actions traceable and separately permissioned. If something goes wrong, you can see that it was the agent, and you can restrict or revoke its access without locking out a person.

It also changes the feel of work. A colleague who never sleeps and sits in the group chat is a new kind of presence. That could be wonderful for tedious tasks. It could also be odd to have a participant who sees all the calendars and approval chains. Companies will need to decide what the agent can see, not just what it can do.

Where you can reach it, and who is trying it

Google says the agent will be available on iOS and Android, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365, ServiceNow and Slack. The reach across Microsoft's own suite is notable. It reinforces the sense that Google wants the agent to follow people into the tools they already use, rather than ask them to move.

Early testers included sportswear brand On, Shopify and PayPal. TechCrunch also lists Gemini Enterprise customers such as BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers. Those are the customers of the broader platform, not necessarily users of this new agent, so I wouldn't read the list as proof of results. The announcement as reported doesn't include performance figures or customer outcomes.

Keeping the bill under control

Agents that plan, delegate and call multiple models can burn through resources quickly. Google addressed that with new flexible spending options, including multi-model orchestration, smart routing and real-time spend caps. The idea is to let businesses keep AI costs under better control.

It's easy to skip past this, but it may decide whether agents become routine. A system that wanders off on a long chain of subtasks can generate surprise costs. A hard cap is the kind of unglamorous feature that makes a finance department say yes.

What to watch, and what remains unclear

A few honest caveats. First, this is a launch announcement, and the coverage I've drawn on is based on what Google said at its own event. How well the agent handles messy, real-world work is still to be seen. Second, the details on pricing, availability dates and the consumer version weren't spelled out in the reporting I read.

Third, and most important to me, is the human side. When an agent can read the calendars, approve-chains and projects of a whole company, questions of accountability get sharper. Who reviews its work? Who is responsible when it makes a mistake? Does it take the dull parts of a job and free people for better work, or does it quietly reshape which jobs exist? The audit trail is a good start. It is not a substitute for clear policies.

If you want to see how other companies are rethinking the interface between people and AI, our piece on GPT-6 and intelligent UI covers a parallel shift. And for the broader workplace picture, our look at AI-created virtual workspaces explores another way AI is changing where and how we work.

Your turn: would you hire it?

The honest question behind this launch is simple. If an AI showed up in your team chat with its own email address, would you trust it with an objective? I would, for small, well-defined jobs with a visible trail and a human checking the result. I'd be warier of giving it the keys to everything at once.

Whatever your answer, this is the direction the tools are heading. The assistant that waits for your question is giving way to one that takes on your to-do list. Think about which tasks in your own day you'd happily hand over, which you never would, and who you'd want watching over the handoff.