Google Just Gave Its AI Agent a Gmail Address. Every Big Player Has Now Built an AI Employee.
On Thursday, October 8, Google Cloud CEO Thomas Kurian walked on stage at Gemini at Work 2026 and introduced what he called "a single, universal agent for work." Most of the coverage focused on the universal part. The interesting part was a smaller feature tucked inside it: coworker agents.
A manager describes a role. Gemini sets up an agent for it. That agent gets its own Workspace account, its own email address, its own calendar, its own Drive storage and its own entry in the company directory. Futurum reports the addresses look like @agents.company.com. It acts under its own identity, not under the identity of the person who created it, and every action it takes is logged to the agent rather than to a human.
Read that list again slowly. Email, calendar, files, a name in the directory, an audit trail with its name on it. That is not a chatbot feature. That is an onboarding checklist.
Four companies, one job description
Google is not alone, and that is the real story. In roughly two weeks, every big player shipped the same idea under a different label:
- Microsoft announced Copilot Autopilot on September 25: an agent with its own Entra identity, audit logs, a role and goals you name, and human sign-off for sensitive actions. Restricted preview, billed in Copilot Credits.
- OpenAI shipped its specialist agent at the start of October, which we covered in OpenAI Shipped an AI Employee. It Costs $200 a Month and You Get Exactly One.
- Manus raised more than $500 million and launched Cue, its own always-on agent.
- Google now gives its agent a mailbox and a desk.
Back in May we argued that "AI employee" would be on every org chart by 2027. I expected some pushback on that one. I did not expect the four biggest platforms in the market to settle the argument for us inside a single month.
The vocabulary has converged too. Nobody is selling "an assistant inside your app" any more. They are selling something with an identity, persistence and accountability. Those are the three properties that separate a coworker from a tool.
What Google actually got right
Credit where it is due, because some of this is genuinely well designed.
Identity is cryptographic, not cosmetic. Each coworker agent gets an attested identity stamped into its logs and into any virtual machine it starts. When something goes wrong, you are not reconstructing which human's token the bot borrowed.
It only sees what you share with it. The agent does not inherit broad organisational access. Team members choose what to make available, the same way you would with a new hire on their first day.
It keeps working after you close the laptop. Multi-hour and multi-day jobs run in the cloud. Anyone who has watched a browser-tab agent die at 80% knows why this matters.
It has a budget. Google announced real-time spend caps that pause an agent when it hits the limit, plus what it calls an AI network firewall that all agent traffic passes through.
It is not locked to one model. Jobs are routed across Gemini and Anthropic's Claude today, with other models promised.
If you work at a company that already runs Google Workspace and Gemini Enterprise, this is a serious option, and you should ask your admin about the preview.
The part that does not fit a 12-person business
Now the honest version for everyone else.
It is a private preview for enterprises. There is no general availability date. Futurum notes that Google did not disclose pricing or GA dates for most of the announcements. A Google representative said there is no intent to charge a per-seat licence for coworker agents in Workspace, which is encouraging, but "no per-seat licence" is not the same thing as "free," and the bill for the work the agent does still has to land somewhere.
It lives inside one vendor's walls. The coworker agent's identity is a Workspace identity. Its files are in Drive. Its memory, its history and everything it learns about your business sit in Google's cloud under Google's terms. That is fine for some companies and a deal-breaker for others, and it is worth deciding which one you are before the agent has six months of context.
Nobody explained how you fire it. The Next Web made a sharp observation: Google's post explains in detail how permissions are granted and says very little about how they are revoked. Offboarding is the half of HR everyone forgets until they need it.
Regulators just started asking the same questions. On the same day as the launch, the UK Information Commissioner's Office opened a call for evidence on agentic AI, covering accountability, transparency and automated decision-making. Responses are due November 20. Its director of technology regulation, Richard Nevinson, put it plainly: agent autonomy "is not an excuse for poor compliance."
Five questions to ask any AI employee, including ours
Since every vendor now sells a coworker, the useful question is no longer "does it have an email address?" They all will. Ask these instead:
- Can I get one today without an enterprise contract? Preview access and sales calls are a cost in themselves.
- Who owns its memory? If you leave the platform, does the agent's knowledge of your clients leave with you, or stay behind?
- How many can I have? One agent that does everything is a bottleneck. A team needs a sales person, a support person and someone running the inbox, each with their own context.
- What does a busy month cost? Not the licence. The work. Ask for a worst case, not an average.
- How do I take its access away, and prove I did? If there is no clear answer, the agent is not ready to hold your keys.
Where this leaves small businesses
The big platforms have now validated the model we have been building since 2025: an AI employee is a teammate with a name, an inbox, a phone number, a memory and a job, not a feature inside someone else's app. That is good news for everyone who was told this was hype.
The difference is who it is built for. Google's coworker agent is designed to slot into a Fortune 100 directory. Most businesses do not have a directory admin. They have an owner who needs someone to answer the inbox by Monday.
That is the gap Geta.Team fills. Every employee comes with its own email and phone number, unlimited persistent memory, and skills it can create on demand. You can run it self-hosted so your data and its memory stay on your infrastructure, bring your own model keys including Claude, and hire as many specialists as you need instead of one generalist. And you can start today with one AI employee for free, no credit card, no preview waitlist.
The question of whether AI employees are real got answered this month by the biggest companies in the world. The only question left is whose building they work in.
Want to test the most advanced AI employees? Try it here: https://Geta.Team
