Chapter 10: Multi-Agent Coordination & Failure Modes
How multiple agents reconcile disagreements (debate, chairperson moderation), the characteristic failure modes of agentic systems (cascading tool failures, role drift, infinite handoff), and the orchestration patterns that mitigate them.
Chapter Overview
Multi-agent systems amplify both the power and the failure modes of single agents. A debate of three agents converges on better answers than a solo agent — when it converges. When it doesn't, you get unbounded loops, role drift, and a frustrating user-facing experience. This chapter is the practical how-to: how to coordinate multi-agent teams, how to detect and recover from cascading failures, and which orchestration patterns are robust enough for production.
This chapter covers:
- Multi-agent debate — multiple agents argue, a judge picks the consensus
- Chairperson & moderation — explicit role for an arbiter agent
- Failure modes of agentic systems — the taxonomy of what goes wrong
- Cascading tool failures — how a single bad observation poisons N downstream steps
- Orchestration patterns for coordination — circuit breakers, deadlines, fallback chains
Chapter Roadmap
Click any topic to jump in
Multi-Agent Debate
N agents critique each other across rounds. 5–15pp accuracy gain on problems with knowledge but weak commitment.
Chairperson
Explicit moderator role — picks speakers, declares consensus, terminates the discussion.
Failure Modes
Spec drift, tool misuse, hallucinated tools, infinite loops, premature termination — and the multi-agent extras.
Cascading Failures
How one bad observation poisons N downstream steps. Circuit breakers contain the blast radius.
Orchestration Patterns
Step budgets, deadlines, fallback chains, idempotency keys — the production hardening that makes coordination safe.
Multi-agent debate (Du et al., 2023; Liang et al., 2023) runs agents in parallel on the same question, has them critique each other across rounds, and aggregates to a final answer. The technique improves factual accuracy and calibration, often by 5-15 percentage points, on problems where the model has the knowledge but commits to wrong answers under direct prompting.
In this topic
The Debate Protocol
A typical debate has three rounds:
- Independent answer. Each agent answers the question on its own. No cross-talk.
- Cross-examination. Each agent sees the other agents' answers and is asked: 'critique their reasoning. Where might they be wrong?'
- Revision. Each agent revises their answer in light of the other agents' critiques.
After rounds, the answer with the most agent votes wins (or a separate judge model picks). Empirically, rounds with agents is the sweet spot. Diminishing returns kick in at .
Why it works: each agent's first answer is a confident guess. The cross-examination forces them to defend — and good defenders update under good critiques. The aggregate is more robust than any individual.
Diversity of Agents
Debate works best when the agents have different priors:
- Different models (GPT-4 + Claude + Gemini) — each has different training data biases
- Different system prompts (one role-played as a 'skeptical critic,' one as 'supportive advocate')
- Different temperatures (one greedy, one stochastic) for diversity in initial answers
Three identical agents with identical prompts produce three identical answers and do not improve under debate. Variance is a feature.