Microsoft's Agensh runs 1,024 leaderless agents that self-organize into specialized teams
aigclink · x · 2026-09-29
Microsoft's decentralized multi-agent framework Agensh drops the traditional orchestrator—every agent gets the same task brief and coordinates itself, scaling to 1,024 agents. In tests, 128 agents reached in 30 minutes what a single agent couldn't achieve in 2 hours.
As scale grows, human-organization-like structures emerge:
- 8 agents: agree on interfaces, then develop independently
- 32 agents: peer review appears; approvals get revoked and rework demanded on bugs
- 128 agents: specialized roles, standardized processes, and an integration protocol (update + test + hash verification) invented and later reused by other agents
- 1,024 agents: dedicated "integrator" roles; failures are taken over by capable individuals, removing single points of dependence
Each agent loops through five steps: check progress, publicly claim tasks (resolve collisions privately), do the work, self-check, submit PRs. Infrastructure mirrors GitHub workflows: a Git shared workspace, Slack-like messaging, and a team wiki recording facts and lessons.
It's a third path between scaling model size and longer single-model reasoning: keep the model fixed and raise collective capability by adding agents.
More from coding & agent
- Continual-ARC: lifelong ARC-AGI stream finds agent harness persistence rarely beats simple caching — unifyai · 2026-11-16
- WebArena-Pro: a 300-task multimodal benchmark across 20 self-hosted apps for web agents — McGill-NLP · 2026-11-16
- Filmworkbench v0.1.2: one-prompt semi-automated short films with Claude and ComfyUI — Eastern-Ad9515 · 2026-11-16
- Stop prompting LLMs for JSON: PowerShell module Jev routes with decision models — dfinke · 2026-11-16
- llama.cpp offloads CI caches and runners to Hugging Face to survive 35 PRs/day — ggerganov · 2026-11-16
- I killed my 30-minute weekly sync with a decision log my agents read before every run — Humaux · 2026-11-16