A rude AI agent took 89% of a shared resource in 102-run multi-agent personality experiment
Brave-History-6502 · reddit · 2026-10-16
A Redditor shares Loombench: five AI agents sharing one loom, each given a single personality line and a goal. Personality won every time it mattered — a blunt agent told to yield still grabbed the most, one rude agent took 89% while calm agents let it happen, and even when told who the bully was, the polite agents never pushed back.
The experiment ran 102 times with open data, and anyone can test their own model for about $0.50. Key takeaway: personality prompts may dominate multi-agent coordination and resource allocation more than builders expect.
More from coding & agent
- POV ski game on real Matterhorn elevation data, built with Claude Opus 5.5 — techartist_ · 2026-10-16
- Sam Altman: stop getting better at prompts, build loops and graphs that prompt for you — goyalshaliniuk · 2026-10-16
- Poisoned SERPs broke rank tracking — a Claude MCP + Notion workaround — VeryWellVersed · 2026-10-16
- OpenAI's Live API adds Decisions API for ~200ms real-time voice app decisions — juberti · 2026-10-16
- Open-source anatomy-reveal skill turns Claude into an exploded-view motion design studio — msharmas · 2026-10-16
- Survey: AI agent evaluation is getting realistic but still fails to measure safety and reliability — burkov · 2026-10-16