Three papers map multi-agent scaling: logistic growth, Ringelmann effect, and -70% architecture mismatches
burny_tech · x · 2026-10-04
burnytech summarizes three studies on agent scaling, all pointing to the same lesson: stacking agents isn't free.
- MacNet (ICLR 2025): DAG-organized collaboration among 1,000+ agents follows a logistic growth law, with collaborative emergence earlier than neural emergence; irregular topologies win.
- Ringelmann effect: across 44 cells spanning Qwen, Llama, Ministral (7B–32B) and Gemini, individual productivity drops as teams grow; the functional form fits at R²>0.99 everywhere.
- Towards a Science of Scaling Agent Systems: across 260 configs, 6 benchmarks, 5 architectures — coordination yields diminishing returns once single-agent baselines are strong; tool-heavy tasks incur multi-agent overhead; architectures without centralized verification propagate more errors. Performance vs single-agent ranged from +80.8% (decomposable financial reasoning) to -70.0% (sequential planning).
Related event: Multi-Agent Scaling Law Studies Spark Discussion(3 posts)→
More from coding & agent
- Engineer uses Claude Opus 5.5 to run river hydraulics calc with 3D viz, in-browser — anselm · 2026-10-04
- Google's VeriHarness shows agent agreement hides shared errors, gains 6+ points with agentic verifier — omarsar0 · 2026-10-04
- NVIDIA paper: verify-then-execute lifts TerminalBench Pass@1 from 50.0% to 68.0% — dair_ai · 2026-10-04
- Dev builds fully code-generated drivable Mars demo with Claude Opus 5.5: 23.7 hours, ~$259 in API costs — bennash · 2026-10-04
- Dev shows off mostly-autonomous agent cluster: one Orchestrator running Claude Code fleets — aksh_stocks · 2026-10-04
- Codex usage resets expire in under 13 hours if unused — just ask Codex for the exact time — JeremyNguyenPhD · 2026-10-04