Skip to main content
All terms
Patterns

Multi-Agent Debate

Multiple agents argue opposing positions and critique each other to reach a stronger answer.

Definition

Multi-agent debate sets two or more agents against each other on a question or proposal. The agents take opposing positions, present arguments, critique each other's reasoning, and iterate until they converge or the system picks the best-supported conclusion. The back-and-forth tends to surface errors and improve factuality on contentious or high-stakes questions, at the cost of extra model calls.