In multi-agent debate (or group chat), multiple agents operate as peers in a shared thread, proposing approaches and challenging each other’s assumptions.
Why It Matters
Debate-style collaboration can surface blind spots, improve coverage, and increase reliability on ambiguous tasks.
Risks
- Cost and latency scale with number of agents.
- Social failure modes (groupthink, dominance, long loops) need guardrails.