Relic turns multi-agent collaboration into persistent organizational protocols, hitting 79.1% on CooperBench
youjiaxuan · x · 2026-10-08
A team spanning UIUC, Yale, Stanford and others introduces Relic, a framework that converts recurring multi-agent collaboration failures into organization-owned, governed, executable protocols that persist across tasks and member turnover.
Key results:
- 79.1% (371/469) on CooperBench, the best reported peer-structured result vs. 59.1% for released Peer and 74.4% for hierarchical Team
- Across 360 controlled runs on three models, Relic improves complete-contract delivery by +5.71 pp over the B2 baseline (95% CI [2.97, 8.55])
Preprint: arXiv:2609.32965, led by Hongyi Du with Jiaxuan You as corresponding author.
More from coding & agent
- Victor Taelin fixes cache-breaking bug in AI coding harness gist, hits 98%+ cache reuse — carsonfarmer · 2026-10-08
- "Everyone's talking about selling data, but no one's talking about selling skills" — tedddyoweh · 2026-10-08
- Claude Code v2.1.294 fixes prompt hooks that failed to block commands they should — ashwin-ant · 2026-10-08
- Dev Building Runtime Authorization Layer for AI Agents Seeks Real Teams to Sandbox-Test It — Invisible_act1988 · 2026-10-08
- Developer uses Codex for CAD and build docs in physical robot hardware project — OpenAIDevs · 2026-10-08
- cogmemai-mcp launches with 28 MCP tools for persistent cloud memory across Claude Code and Cursor — modelcontextprotocol · 2026-10-08