I Don’t Trust Any One AI. So I Make Them Audit Each Other.
Here is the problem with asking one AI a question: it answers like it’s sure. Every model does. Confident tone, tidy structure, no hesitation — whether it’s right or completely wrong. If you take the first answer on its word, you haven’t used a tool. You’ve adopted an opinion you can’t defend.
So I don’t take any single AI’s output on its word. I run a tribunal.
How the tribunal works
It’s a method, not a gadget — the same discipline, applied every time the stakes are real:
- One AI does the work. A draft, a plan, a check, a recommendation — whatever the task needs.
- Other AIs cross-examine it. The work gets handed to different models to audit: where is this weak, what did it miss, what did it assume? Gemini sits as an auditor among them — that’s the job I use it for.
- Disagreement is the useful part. Agreement is a signal, not proof. What survives the argument gets checked against the primary source — and the final call is mine. When they don’t, I learn exactly where to dig before I act.
- I make the final call. The tribunal advises. It does not decide. The verdict is mine, the same as it’s always been.
No single AI gets taken on its word. Not even mine.
Especially not mine. The systems I rely on every day are the ones that earn the hardest audit — because I’m the one who acts on what they say.
Why I audit my own systems hardest
People assume checking is something you do to strangers — a seller, a vendor, a listing. I do it to my own tools first. My agent builds my websites, checks my prices, and helps plan my routes. If she’s wrong and I don’t catch it, I’m the one standing in the wrong RV park, or shipping the wrong page, or spending the wrong money.
You’ve already seen the small version of this on this site: when Claude planned my Highway 101 route, I made it re-check its own work against my two rules — and it caught one of its own misses. The tribunal is that same instinct, scaled up: don’t just re-read it yourself. Make other systems, with different blind spots, try to break it.
The point
This is the discipline I now point at vendors for a living. If I audit my own AI systems with a tribunal — systems I chose, that work for me, that I trust enough to run my days through — imagine what I’ll ask on your sales call, when the system was built by the people trying to sell it to you.
Kalyn Wright runs her businesses from the road, full-time in a 42′ fifth wheel since May 2021 — and trusts no AI output, including her own systems’, until the tribunal has had its say and she has made the call.