OpenAI’s own guidance is blunt: shorter prompts beat long instructions, outcome-first beats step-by-step. This is relevant, because every model release moves further in that direction. As models improve, they reward cleaner briefs. The skill that compounds is the brief.