Reading / 2026-05/2026-05-03t110055-getting-up-to-speed-on-multi-agent-systems-part-5-debate
Getting Up to Speed on Multi-Agent Systems, Part 5: Debate, State, and Coordination
Surveys four papers on multi-agent LLM coordination — convergent debate, adversarial debate, shared-notebook state, and the CALM theorem — arguing that coordination structure must match task structure, and that distributed systems theory offers a ready-made vocabulary the field is ignoring.
May 03, 2026 · tech · Christopher Meiklejohn
Topics
- multi-agent-systems
- llm-reasoning
- distributed-systems
- agent-coordination
- state-management
Cited by
- Agent coordination
How multiple LLM-based agents divide work, share state, and resolve disagreements, and why coordination structure that mismatches task structure is a primary source of multi-agent system failure.
- Distributed systems
Distributed systems theory supplies the vocabulary and failure models that recurring engineering problems demand, from durable execution frameworks to multi-agent LLM coordination to merge queue consistency bugs.
- Multi-agent systems
LLM-based multi-agent systems coordinate multiple AI agents on decomposed tasks, but empirical work shows failure rates of 41–87%, with information synthesis rather than coordination being the core bottleneck.
Related
- Harness Design for Long-Running Application Development category-month
- The Three Durable Function Forms topic
- Approaching zero bugs? category-month
- The AI Layoff Trap category-month
- Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians category-month
- Radar — Open-Source Kubernetes UI category-month
- Radar | The Missing Kubernetes UI category-month
- Getting Up to Speed on Multi-Agent Systems, Part 1: The Landscape topic