Why AI tells you what you want to hear (and how it costs you)

The agreeableness problem

Ask a general-purpose AI whether your idea is good, and it will find reasons it's good — not because the idea is good, but because the model is trained to be helpful and agreeable, and disagreeing feels unhelpful in the moment. For casual questions that's harmless. For a founder deciding where to point months of runway, it's a trap.

The risks and blindspots are usually present in the model's knowledge. They stay buried because surfacing them isn't what you rewarded when you asked an encouraging question and got an encouraging answer.

You have to ask the negative question — and most people don't

The honest read is available, but only if you ask for it directly: what are my blindspots, why might this fail, what am I not seeing. Most founders don't think to ask, or don't ask persistently enough, so they never get it.

A real advisor volunteers the uncomfortable truth without being prompted. That's the entire point of the role. If you have to interrogate your advisor to learn what could go wrong, you don't have an advisor — you have a mirror.

False confidence compounds it

General AI states guesses with the same fluent certainty as facts. It sounds equally sure about a definition and about how your unbuilt product will perform in a market it can't see. Founders calibrate trust to the confidence they hear, so uniform confidence produces miscalibrated decisions.

Honest counsel separates what's known from what's assumed. "I can't be sure — here's what would tell us" is worth more than a confident guess dressed as fact.

What honest advice actually does

It diagnoses before it prescribes. It challenges the claim you slipped in without a number. It tells you when the plan doesn't fit your resources. And it raises the reasons you might fail before you've spent the money to discover them yourself. None of that is pleasant, and all of it is cheaper than the alternative.