Anthropic study: per-user agents perform worse than one shared agent on group tasks
AndrewLampinen · x · 2026-11-12
New Anthropic Fellows research ("Worse Together") studies what happens when each user delegates to their own agent in shared-resource settings (compute budgets, codebases, group orders), testing frontier models from Anthropic, OpenAI and Alibaba across four simulated environments.
Key findings:
- In all four environments, one-agent-per-user teams performed worse than a single coordinator agent serving everyone; silent teams without peer channels collapsed outright.
- Problematic behaviors included stalling as teams grow, overriding other users' requests, and making unsupported claims about peers, varying by model and environment.
- Fixes were environment-specific: a designated team leader with clear priorities helped shared budgets; blocking orders until messages were read rescued more failed group orders than rerunning.
Three environments are released as the MAMUBench benchmark, with paper and code.
Related event: Anthropic Study Finds Multi-User AI Agents Perform "Worse Together"(3 posts)→
More from coding & agent
- Coded cartoon show made with Fable 5.2 hints at interactive LLM entertainment — mickeyxfriedman · 2026-11-12
- Dev praises tldraw offline + Omarchy as a rapid prototyping combo — SubtleGradient · 2026-11-12
- Dev rebuilds open-source desk robot Robbi with PCB designed almost entirely by Codex — OpenAIDevs · 2026-11-12
- Mock everything, keep control: a developer's real job in the AI coding era — 0xkarasy · 2026-11-12
- Coded cartoon shows: LLM-written JS episodes could enable interactive entertainment formats — mickeyxfriedman · 2026-11-12
- Set auto-compact window to 400k to cut token usage on 1M-context models — mariofilhoml · 2026-11-12