6 min read
What happens to your business when the AI tool you use stops borrowing your login and gets one of its own?
That question is no longer hypothetical. In the space of two weeks, the two biggest office software companies described AI “coworkers” that live inside your systems with their own identity, their own account, and their own permissions. Microsoft’s new Autopilot agent has, in the company’s words, its own identity, memory, computer and workspace. Google’s new coworker agents get their own Workspace account, complete with an email address, a calendar, and a Drive.
Both products are aimed at larger companies first and are still rolling out. But the design choice behind them is going to reach small business tools quickly, and it is worth thinking through now, while you still get to set the rules. This is a strategy piece about AI agent permissions: what is changing, what is genuinely better about it, what is overhyped, and the five rules a solo owner should adopt before any agent moves in.
What Microsoft and Google actually announced
On September 25, Microsoft introduced a redesigned Copilot with Home, Code, and Autopilot. Autopilot, previously called Scout, is a persistent agent that can monitor channels, follow up on threads, run recurring tasks, and pick projects back up days later without a new prompt. Microsoft says it runs with permissions, audit and governance behind it, and that you set its objective and boundaries. It was headed to private preview at the end of September, and Microsoft says this agent work is billed by usage.
On October 8, Google followed with its Gemini at Work announcements. You describe a role, and Gemini creates a coworker agent with its own Workspace account. Google says the agent acts under its own identity rather than yours, sees only what you or your team share with it, and that every action it takes is written to an audit trail attributed to the agent, not to a person. Administrators can also set a hard spending cap per project, and the agent pauses when it hits that cap.
Why this is genuinely better than the old way
Here is the part the skeptic in me has to admit: giving an agent its own account is a safer design than what most solo owners do today.
Right now, the typical small business AI setup works by handing a tool your own access. You connect it to your Gmail, your calendar, your accounting app, all under your name. When something goes wrong, there is no clean way to tell what you did and what the AI did, and removing its access often means untangling several connections.
A separate agent identity fixes three real problems:
- Attribution. If a strange email goes out or a file gets edited, the log says the agent did it, not you.
- Scope. The agent sees only what you deliberately share with it, instead of everything your account can reach.
- The off switch. Turning off one account is far easier than revoking a dozen app connections.
Add Google’s spending cap and you have the ingredients of a sensible arrangement: a helper with a badge, a limited set of keys, a budget, and a record of everything it touched.
What is overhyped for a one-person business
Now the skeptical half. The marketing language of “digital coworkers” glosses over a few things that matter more when there is no IT department behind you.
First, a log only helps if someone reads it. Big companies have administrators watching audit trails. You have a Tuesday afternoon between client calls. An agent that makes a mistake at 2 a.m. and records it faithfully has still made the mistake.
Second, “its own account” can quietly become “its own access to everything.” The safety comes from sharing only what the agent needs. In practice, it is tempting to share your whole Drive because picking folders is tedious. Do that and you have recreated the old problem with an extra email address.
Third, usage-based billing is hard to predict. Microsoft bills this agent work by usage, and Google’s own answer to runaway agent costs is a spending cap. That is a sign that costs can climb. If an agent works while you sleep, it can also spend while you sleep.
Fourth, availability. Neither product is a small business plan you can sign up for this afternoon. The question is not “should I buy this” but “what rules should I already have when tools like this reach the apps I use.”
Five rules to adopt before any AI agent gets an account
1. One agent, one job, written down
Before you create or connect any agent, write one sentence describing its job and what “done” looks like. “Draft replies to new booking inquiries and leave them in a drafts folder” is a job. “Help with my email” is not. We argued this in detail in our piece on always-on agents getting one job, not ten.
2. Share folders, never the whole drive
Create a dedicated folder or label for the agent’s work and share only that. If the job later needs more, add access one piece at a time. Least access first, expand only when you hit a wall.
3. Drafts before sends
For anything that leaves your business (emails, invoices, posts, payments), the agent prepares and you approve. Keep it that way for at least the first month, and permanently for anything involving money. The same logic runs through our take on letting AI draft payroll without trusting it blindly.
4. A budget cap you set yourself
If the tool offers a spending limit, set it low and raise it later. If it does not, set a calendar reminder to check usage weekly for the first month. A surprise bill is the most common way small businesses get burned by usage-based AI.
5. A ten-minute weekly log review
Once a week, skim what the agent did. You are not auditing every line; you are looking for anything you did not expect. If you never find surprises after a month, you can loosen up. If you do, tighten scope before expanding anything. Our one-page AI policy guide is a good place to write these rules down.
Where this is heading, and what to prioritize
The direction is clear: AI tools are moving from “assistant you talk to” toward “helper with its own seat in your systems.” For a solo owner, that shift is neither something to fear nor something to rush into. The separate-identity design is a real improvement, but the safeguards built for companies with IT teams only protect you if you do the small amount of setup they assume.
So prioritize habits over products. Get used to giving AI narrow jobs, limited folders, approval steps, and a budget now, with the tools you already use. When agent accounts arrive in the small business plans you pay for, you will already know exactly what to share and what to keep.
Which job in your business would you trust to an agent with its own account, and which would you never hand over? Tell us in the comments.
Related reading
- Always-On AI Agents Like OpenAI’s Dots Should Get One Job From a Solo Business, Not Ten
- Only 23% of Small Businesses Have an AI Policy. 9 Quick Wins to Write Yours on One Page
- QuickBooks AI Can Now Draft Your Payroll. Let It Do the Typing, Never the Trusting
- Notion’s AI Agents Now Work for 24 Hours and Connect to 150+ Apps: A Solo Owner’s Spotlight
