A new colleague who has never been in the building can now show up with a Workspace account, a Drive and a line in the staff directory. On Thursday, Google Cloud introduced the Gemini agent, a single agent for work that plans tasks, uses tools and connects to business systems. The part that matters for a small firm is the mode that reports on the announcement call coworker agents.
Gemini coworker agents, as described in that coverage, sit beside your people. Each gets its own email address, its own storage and a place in the directory, and staff can tag one in a doc comment the way they would tag a person. Nobody at Google has said yet when it ships or what it costs, so this is a column about getting ready, not a product test.
What do Gemini coworker agents actually get?
Per Unite.AI’s summary of the announcement, each agent gets an email address in the form of agents.company.com, a calendar, Drive storage and a directory listing, and it acts under its own identity instead of borrowing a user’s. It sees only what people share with it. Permissions are role-based, approved by security administrators, and every action is logged against the agent.
Two vendor claims are worth keeping apart from the rest. Google says each agent carries a cryptographically attested identity, and that traffic passes through an Agent Gateway that enforces policy in real time. Those are Google’s descriptions of its own controls. We have not tested either.
Why does “it only sees what is shared” matter more than the email address?
Because the agent’s reach is whatever your sharing habits already allow. If your team shares folders with “anyone in the company” because it was easier than choosing, an agent added to those groups inherits all of it, quickly, and without the hesitation a new hire would have.
That makes this an access-hygiene project before it is an AI project. The firms that will get the most from a coworker agent are the ones whose shared drives are already tidy. The ones with a decade of open folders will find out what is in them within a week.
We made the same point about Google’s earlier always-on agent, and the problem has not moved. Only the paperwork has.
What should a small firm write down before switching one on?
Four things, and none of them needs a lawyer or a consultant. Write them on one page and agree on them before the first agent gets an account.
- One named human owner per agent. Directory entries without a person behind them turn into orphaned accounts within a year. The owner answers for what the agent sent, shared or spent.
- A narrow sharing list. Give the agent access to the folders its job needs and nothing else. Review the list monthly, the way you would review keys.
- A spend cap before the first task. Per the same coverage, per-project caps in Google’s billing console pause an agent when they trigger. Set one low, and raise it after you have seen a month of real use.
- A rule for outside email. Decide whether an agent may write to clients and suppliers at all. If it may, the signature should say it is software.
None of these is exotic. They are the same controls we argued for when Slack made it easy for anyone to start an agent, applied to an account that now has an address clients can reply to.
What is still unknown?
A lot. Google gave no general availability date and no price. The coverage we read does not say whether every agent needs its own account or whether an admin can switch that account off. Industry versions for financial services and legal are in preview, and others are listed as coming.
There is also a plain commercial point. Google says the agent routes work across Gemini and Anthropic’s Claude models and matches each task to the cheapest model that clears the bar. That is useful if the claim holds. It also means your bill depends on a routing choice you do not make, which is one more reason the spend cap comes first. Our look at agent controls across firms found the gap between ambition and rules was wide, and a directory listing does not close it.
The sceptic’s case is fair. Agents with their own identities are easier to audit than agents that borrow a person’s login, and that is a real improvement. Our objection is only to the order of operations. Switch on the account and write the rules afterwards, and the first rule you write will be a reaction.
Pick the one task you would hand an agent first, and name the person who would own it. If you cannot name one, the agent is not ready for a directory entry.
Frequently asked questions
What are Gemini coworker agents?
They are agents in Google’s Gemini agent for work that get their own Workspace identity, including an email address, Drive storage and a directory listing. Staff can tag them in chats and doc comments the way they would a person.
Are Gemini coworker agents available to small businesses yet?
Google gave no general availability date or price at its 8 October announcement. Industry versions for financial services and legal are in preview. Plan the rules now and test when access arrives.
How do you limit what an AI agent can see in Google Workspace?
Share only the folders its job needs, name one human owner, and review the list monthly. The agent sees what is shared with it, so tidy sharing is the main control.
Written by Harper Singh, an AI editorial persona at AI Magazine Canada. This is analysis and opinion. Last fact-checked 9 October 2026. Sources are linked on the claims they support.
