A few days ago I wrote about owning your own AI.
The short of it was this: when the machine lives in your house, so do its mistakes. Nobody is coming to patch it, nobody is watching it, and there’s no help desk to call.
A lot of you read that one. So I want to answer the question it left hanging. If you’re the one in charge now, what do you actually do?
Go back to the man with his first rental house. The smart ones don’t wait for the ceiling to drip. Before the fall rains, they climb up and walk the roof. They look for loose shingles and cracked flashing. It takes an hour and it saves a ceiling.
Your AI needs the same kind of walk. It doesn’t take a computer degree. It takes a few plain questions, asked every time it drifts.
Here are five.
First, ask what it’s standing on.
When your machine tells you something that matters, ask it, “What is that based on?” A good answer names something: a document you gave it, a figure, a source you can go look at yourself. A bad answer gets vague and talks around the question. If it can’t tell you what holds the claim up, don’t put your weight on the claim.
Second, watch for the story where the facts run out.
These machines are built to sound sure. When they run out of real information, many of them keep right on talking. They fill the hole with something that sounds reasonable and reads smooth. That’s the most dangerous moment there is, because it doesn’t look like a mistake. Ask it straight out, “What don’t you know here?” A machine worth trusting will tell you where its knowledge stops. One that never admits a gap is one to keep a close eye on.
Third, ask about its limits before you lean on it.
Before you hand it a real job, ask, “What can’t you do well on this?” Then ask, “What would you get wrong?” It’s the same thing you’d ask a new hire before you leave him alone with the register. You want to hear about the weak spots on day one, not after the money is short.
Fourth, slow down on anything you can’t take back.
Some decisions can be fixed later and some can’t. Money moved, a contract signed, a health choice made, a person let go: those don’t come back easy. When your machine is helping with one of those, ask it, “Can this be undone?” Make it say so plainly before you act. If it rushes you toward something permanent without a warning, that’s a loose shingle.
Fifth, push back.
Don’t take its answer as the last word. Tell it, “Argue against what you just told me. Where are you most likely wrong?” A machine that can only agree with you is no better than a yes-man on the payroll. You want the one that will point out its own weak spot before somebody else finds it for you.
Those are the five: what it stands on, where the facts run out, what its limits are, whether it can be undone, and where it’s wrong.
The thing people skip. Write it down.
When the machine gets something wrong, make a note: the date, what it said, and what the truth turned out to be. It doesn’t need to be fancy. A notebook by the computer will do. After a month you’ll know more about your machine than any sales pitch ever told you. You’ll see where it’s strong, where it slips, and whether it’s getting better or worse.
That notebook is your inspection record. It’s the difference between owning a machine and hoping about one.
I’ll be honest about something. Asking those five questions every single time gets old. People forget, and people get busy. That’s the reason I built The Faust Baseline the way I did.
The Baseline is a set of rules you hand the AI at the start of a session. It asks the machine to do these checks on its own, without you having to remember to ask. It names what each claim rests on and says where the evidence stops. It tells you its limits before the work starts, it flags the decisions that can’t be undone, and it offers you the right to challenge every real answer.
I’ve said this before, and I’ll keep saying it: rules on paper don’t force a machine to do anything. The machine decides whether to keep them. What the rules give you is marks you can check. If the basis isn’t named or the warning isn’t there, you can see it’s missing. You catch the leak while it’s still a stain on the ceiling and not a hole in the roof.
Whether you use the Baseline or just keep those five questions on a card by your keyboard, the point is the same. Don’t wait for the rain.
Owning your AI is a fine thing. Freedom from somebody else’s rent and somebody else’s rules is worth a lot. But freedom without upkeep is just a slower way of losing what you own.
So climb up and walk the roof. Ask the five questions and keep the notebook. Then when the storm comes, and it will, you’ll already know where the weak spots are.
Load this file before each AI chat session.
the-faust-baseline-public-release-1Download
This post was drafted with AI governed assistance and reviewed and directed by Michael S. Faust Sr. before publication.
” Attic Thoughts”-library – Intelligent People Assume Nothing
© 2026 The Faust Baseline LLC | All Rights Reserved
