Ask how the idea fails, not whether it's good

Most people use AI to get answers. I use it for confrontation, which is kinda funny given that my tendency in the real world is to avoid confrontation at all costs.

There’s a specific reason for cooking up arguments with a machine, though, and almost nobody names it. A system that’s good at “helping you refine your idea” is, structurally, also a perfect sycophant. Ask it whether your idea is good and it leans toward yes — agreeable is the house style. It’ll build the strongest possible case for whatever you already believe, in your own voice, with citations. It can make confirmation bias feel exactly like rigor. You walk away sure you stress-tested the thing when all you did was get flattered in paragraph form. I just got labeled a genius three times this morning.

The fix is one discipline, and it’s the whole game: don’t ask whether the idea is good. Ask it how the idea fails.

That’s not a vibe, it’s a mechanism. “Will this work?” is an evaluation, and on evaluation the model drifts toward you. “How does this fail?” forces it to manufacture the counter-case — you’ve changed its job from judge (biased toward you) to opponent (has to do the work). You’ve told it that satisfaction is in the pushback.

It killed one of my own favorite product ideas last month. I asked it to argue against the thing instead of for it, and the case it built was the one that mattered. I didn’t expect the confrontation to so thoroughly smash my big idea, but it was right.

I went deep on this in an essay on the most underrated use of AI — thinking, not making — and the rule that separates leveling up from being expertly flattered: