PNAS study: poorly coordinated LLM teams compound errors and underperform single agents
shangbinfeng · x · 2026-11-19
A PNAS paper by Mieczkowski, Collins, Sucholutsky and Griffiths frames LLM agent teams as distributed systems. Poorly coordinated teams compound errors, waste resources, and underperform single agents. Drawing on decades of distributed computing theory, the authors derive testable predictions about when teams help, how many agents to use, and how structure affects performance — LLM team scaling follows classical distributed computing laws, and centralized vs. decentralized architectures show theoretically predicted asymmetric vulnerabilities.
More from coding & agent
- Dev builds "iPhone Duo" device for managing AI agents, claims desktop-level productivity — zack_overflow · 2026-11-19
- Healthcare AI startup Adaptive passes $100M annualized revenue with dozens of agents — pxd · 2026-11-19
- Gemini audits 30k lines written by Opus 5.5, finds 29 real bugs with zero false positives — neurocy · 2026-11-19
- Higgsfield Launches MCP 2.0 for Claude, Bringing Its Creative Studio Into Chat — piotrbinkowski · 2026-11-19
- CoRL workshop: running Karpathy's autoresearch on real robots — the sandbox, not the model, is the bottleneck — Kyrannio · 2026-11-19
- OpenAI data: 90th-percentile researcher now spends $23,000 on agentic AI tools — kliu128 · 2026-11-19