Multi-agent training spontaneously evolves middle management with 'lead' roles
rosstaylor90 · x · 2026-10-23
rosstaylor90 shares early findings from a multi-agent organization study on how agent teams divide labor:
- Hierarchies emerge spontaneously: teams start as a wheel (an imposed CEO agent connected to everyone), but over training CEOs increasingly hire "Lead x" roles, creating a middle-management layer (charts show lead engineers rising from 0 to 11%+ of hires).
- Matches organizational economics theory: communication costs and information aggregation predict hierarchies; agents rediscovered this via collective reward.
- Leads aren't just summarizing: they are the chief builders, with teams organizing verification/QA support roles around them — like the "surgical team" model from The Mythical Man-Month.
- Method: multi-agent credit assignment is hard (Kimi K2.5 froze subagents and only trained the orchestrator); the team tweaked a CISPO-like recipe to use a flat collective advantage per agent, and it works surprisingly well.
More from coding & agent
- Cursor agents can now drive your desktop, and one dev warns self-verifying agents just move the quality gate — petrusenko_max · 2026-10-24
- End of 'good enough software': formal verification becomes the missing AI demand curve — bookwormengr · 2026-10-24
- Dev says agents now run everything in his workflow except scrolling Shorts — tokenbender · 2026-10-24
- DAB, a cross-database agent benchmark, hits EMNLP Oral; best model only 38% pass@1 — adityagp · 2026-10-24
- Firecrawl Alexandria launches sports data APIs across 14 providers after $75M Series B — devdigest · 2026-10-24
- 27B open-source model beats GPT-5.5 and Opus 4.8 on desktop tasks at 1.5% of the cost — alex_verem · 2026-10-23