One Solopreneur's Rules for What AI Can and Cannot Touch

A solo business owner lets AI read his QuickBooks financials every month—but the same AI cannot edit a single transaction. That split is the core of a permission system he says lets him use AI heavily without putting his business at risk.

The owner, writing in a first-person account published by Fast Company, argues that most AI advice skips the hard question. Connecting an AI tool to an account isn't just a convenience decision. "Every time you connect an AI tool to one of your accounts, you're also answering a second question: What is this tool allowed to do on my behalf?" he writes.

For many solopreneurs, that question gets answered by default. You click "allow" on the permissions screen, and the tool can now reach everything in that account. He treats that as a mistake and sorts every AI tool he uses into three buckets.

What Are the Three Access Buckets?

Read but not edit. AI can pull information from the tool but can't change anything. The owner uses this for reference material he wants AI to draw from without any risk of it being rewritten. His example: pulling financial data from QuickBooks, where Claude analyzes numbers monthly while transactions stay off-limits to machine edits.

Full editing. AI can create and change things directly. He grants this only where a mistake is easy to spot and quick to undo. He plans marketing material in Canva, calls nothing there mission-critical, and gives Claude full change capabilities inside the design tool.

Off-limits. AI can't access the tool at all. His bank accounts sit firmly in this bucket. His reasoning has an external data point behind it: JP Morgan Chase reported, per CNBC, that people were handing over their banking credentials to AI agents like Muse. "That's outside of my comfort zone," he writes.

Can Partial Access Replace All-or-Nothing?

Some of his most useful setups give AI partial access rather than none. He describes three workarounds.

Secondary accounts. Give AI its own limited version of a tool and keep it out of the main account. He runs a secondary email inbox for inbound inquiries to his website. AI sorts messages and drafts replies there without ever touching his primary inbox.

Limited permissions. Some tools let you scope AI to specific projects. He manages business information in Airtable, and Claude can only access certain parts of the database. That way it can't inadvertently cause an issue in a tool he relies on heavily.

Supervised tasks. AI does the work, and nothing moves forward until he approves it. In his email setup, Claude drafts a reply but can't send it. The draft waits for his review.

Why Do the Boundaries Need Revisiting?

The owner treats his rules as provisional. AI tools keep adding new connections and capabilities, and those changes may make a user more or less comfortable over time.

He offers his own email history as proof. He was resistant to letting AI access his inbox at all. But the inbound inquiries he receives from other people using AI themselves forced him to think differently—his phrase for it is "use AI to fight AI." His solution was the separate email inbox, which gave AI a contained workspace instead of the keys to his main channel.

His closing claim is a governance argument, not a productivity one: "In the end, I can use AI as much as I do because I decided up front what it can access." The heavy usage and the strict boundaries, in his telling, aren't in tension—they're the same decision.

What's the Practical Takeaway for Small Operators?

The system costs little to copy. It requires no new software, only an audit of which tools AI already reaches and a decision—made deliberately rather than by default—about whether each connection belongs in the read-only, full-edit, or off-limits bucket.

The banking example sets the outer limit. Chase's reported findings on users sharing credentials with agents like Muse suggest the default-permission problem is already widespread, not theoretical. For solopreneurs who run their entire operation through a handful of SaaS accounts, the permissions screen has quietly become a business-risk decision, and the writer's advice is to treat it that way before clicking allow.