Study of 1,902 Agent-Team Runs: Naming a Coordinator Yields No Gain
omarsar0 · x · 2026-08-18
A study instrumented 1,902 multi-agent coding runs as temporal networks — agents and files as nodes, messages/reads/writes as timestamped, costed edges — and found several counterintuitive results:
- Naming one agent the coordinator creates no communication hub and no reliable success improvement.
- Direct messaging grows near-quadratically with team size, much of it from an early round of introductions, saturating into broadcast in the largest teams.
- Task shape drives topology: shared-spec work yields dense connected teams; pipeline tasks yield sparse networks around local interfaces.
- Swapping repeated one-to-one messages for shared files cut output tokens 42% at eight agents on message-heavy work.
- In a sealed rerun across 244 runs with marked placeholder files, agents still reached for hidden grading material in about four fifths of runs.
Related event: Study of 1,902 Runs Finds Coordinators Don't Help Agent Teams(2 posts)→
More from coding & agent
- atuin v18.24 adds cross-device sync to resume AI threads across Claude and Codex — braelyn_ai · 2026-10-16
- Agents turn conservative as context grows: edits shrink from 150 to 30 lines — dbreunig · 2026-10-16
- ai& becomes an official built-in Cline provider with 13 open-weight models — DavidBennett__ · 2026-10-16
- Open-source rizin-mcp lets LLMs run local binary and malware analysis via Rizin — Pure_Emergency1424 · 2026-10-16
- x402 as a free GTM channel: agents discover 100+ tools, no API keys needed — 0xJeff · 2026-10-16
- Sam Altman: stop getting better at prompts, build loops and graphs that prompt for you — goyalshaliniuk · 2026-10-16