Chain-of-Thought (CoT) prompting elicits step-by-step reasoning from LLMs, dramatically improving performance on complex reasoning, math, and logic tasks. Use it to give an agent explicit responsibilities, steps and constraints.