Anthropic Study Finds Multi-User AI Agents Perform "Worse Together"

Anthropic Fellows' new study "Worse Together" shows that when multiple users each deploy their own agents to share resources, multi-user multi-agent coordination often breaks down and performs worse. The team built four scenarios and open-sourced the MAMUBench benchmark.

2026-11-12 ~ 2026-11-12 · 3 related posts

• Anthropic study: per-user agents perform worse than one shared agent on group tasks — AndrewLampinen · 2026-11-12
• Retweet: Anthropic Fellows research on multi-user multi-agent coordination — AndrewLampinen · 2026-11-12

1 near-duplicate retellings: AndrewLampinen