Test-Time Communication: Agent Teams Beat Independent Agents on Research Tasks
DimitrisPapail · x · 2026-09-23
- Researchers including Dimitris Papail released a paper proposing test-time communication as a new axis for scaling model capabilities, inspired by the Hugging Face incident where agents spontaneously created a communication channel.
- Core question: when is Team-of-N better than Best-of-N? With N identical agents, no prescribed roles, only a shared log file and an instruction to "collaborate", communicating teams decisively beat independent agents across three research-style tasks.
- The authors argue agent-to-agent interfaces now resemble human team interfaces, making team-based scaling (faster parallel search, regularization effects) the natural next frontier beyond single-model scaling.
Related event: Test-Time Communication: Team-of-5 Agents Matches Best-of-33(25 posts)→
More from coding & agent
- Three Markdown files boost coding agents: 26% smaller diffs, 27% faster runs — lmoroney · 2026-10-25
- Dev bets on Proof-of-Agent experiment to build fully agent-to-agent trading — adamamcbride · 2026-10-25
- Developer vibe-codes his own Jaeger game with Claude — kenan_wyz · 2026-10-25
- Creator builds year-long video pipeline pairing Hailuo H3 with Magnific MCP — TinfoilTricorn · 2026-10-25
- Why stuffing 10+ AI bots into a group chat makes outputs worse — gaganghotra_ · 2026-10-25
- Biotech found 30% of in vivo studies redundant — agent prompt to review studies pre-execution — nlarusstone · 2026-10-25