Don Lovett, Fractional CFO & Managing Principal · ProjectBits Consulting · July 2026

~4 minute read

The decision that determines whether AI keeps you in control is not which model you pick — it’s whether your process is documented, the AI’s boundaries are named, and your business logic lives on a page you own rather than inside a model you can’t open. All three fit in a 20-minute, one-page exercise you can do this week. This is the activation companion to The Owner Who Answered First.

In part one, I told you about an owner who was completely insulated when the $10B–$100B copyright lawsuit hit Google’s Gemini — not because she picked the right model, but because she’d answered the real question a year in advance: whose intelligence, obtained how, and what happens to my business if it changes?

The reaction I get to that story is always the same: "Great for her. She had time and a consultant. I have a business to run."

So here’s the honest version. You don’t need eighteen months or a consultant to start being that owner. You need about twenty minutes and the willingness to write three things down. This is the move. Do it this week.

The 20 minutes

Minutes 1–8: Write down one process the way it actually happens.

Notice the direction here, because it’s the whole discipline in one instruction: start from a business problem, not from the AI. The vendor’s question is "here’s the model — where can we use it?" Yours is the opposite: "what actually hurts in my business, and what would fix it?" Pick the single workflow you’d least want an AI to get wrong — how a client gets onboarded, how a reconciliation exception gets handled, how a job gets priced — because it’s a real problem, not because it’s a place AI could go. The fix might turn out to be a checklist, a clearer role, or AI. You decide that after you understand the problem, never before. Write the process as if you were training a new hire who will take it over tomorrow. Include the judgment calls: when do you override the default? what’s the edge case everyone forgets?

This is the whole ballgame, and here’s why. Most businesses run on tribal knowledge — logic that lives in your head, invisible to anyone who wasn’t there when it formed. When you hand an undocumented process to AI, it doesn’t automate your process. It invents one. The output looks right. It works most of the time. But the logic underneath is the AI’s logic, not yours — and that gap is the very first place agency drains away. You cannot govern what you cannot see. Eight minutes of writing turns a dependency back into a workflow you own.

Minutes 9–15: Draw three boxes for the AI.

On the same page, write three headers and fill them in for that workflow:

• Green — the AI decides on its own. (e.g., "drafts the first-pass onboarding email.")
• Yellow — the AI must stop and I review before anything happens. (e.g., "any exception over $500 waits for me.")
• Red — the AI never does this, no matter how confident it sounds. (e.g., "never touches a client’s tax classification.")

This is what "humans above the loop" actually means in practice — not a philosophy, a page. The owner who has named the green, yellow, and red boxes before deployment is driving. The owner who hasn’t has a workflow designed around the vendor’s start date, not their own authority. If those boundaries aren’t written down before the engagement, silent execution is the default: the AI making decisions inside your business that you never named and cannot defend when a client asks.

Minutes 16–20: Write the one sentence that keeps the model upstream.

At the bottom of the page, write: "AI helps me design and document this. A human validates the rules. The business runs on the rules, not on the live model."

That sentence is your insulation. It’s why the owner in part one was untouched by the lawsuit. Her critical logic lived in documented rules she owned — the model was upstream of her workflows, not embedded in them. So when a court reshapes the model (and courts are reshaping the frontier models right now), her month-end close doesn’t change. Yours won’t either, if the logic lives on your page instead of inside a black box you can’t open.

That’s it. You’re now the owner who answered first.

Twenty minutes and one page per critical workflow. You will not have documented your whole business. You will have done something more important: you’ll have proven to yourself that you can — and you’ll have made the exact move that separates the owners who stay in control from the ones who wake up dependent.

Notice what you did not have to do. You didn’t pick a model. You didn’t learn the difference between Gemini and Claude and GPT. You didn’t predict which lawsuit lands next. None of that was ever the decision that determined the outcome. The decision that determines the outcome is whether the context is governed, the process is documented, the boundaries are named, and the audit trail is defined — before anything gets deployed.

The lawsuits will keep coming. The models will keep shifting under everyone who built on top of them. And a small number of owners — the ones who spent twenty minutes with a page — will keep running their businesses on logic they own, judgment they can explain, and answers their clients can rely on.

Be one of them. Start with one page, this week.

Want the full architecture behind the 20-minute move? Read AI Debt: The Tax on Small Business [projectbits.com/insights/ai-debt-tax-on-small-business/], and find out which financial persona is running your business today at [projectbits.com/which-financial-persona/].

Don Lovett is the founder of ProjectBits Consulting, a fractional CFO and bookkeeping firm, and the developer of the CFO Operating System™.