Agent Debate

Also known as: multi-agent debate

Agent debate is a pattern in which two or more agents argue opposing positions on a question, critique each other's reasoning, and converge on a shared conclusion through structured dialogue. The adversarial dynamic surfaces assumptions, identifies weaknesses in arguments, and tends to produce more thoroughly reasoned outputs than a single agent working alone. Debate is especially useful for complex decisions where a single model's bias might otherwise go unchallenged.