Microsoft's Agensh Scales Self-Organized Multi-Agent Coding to 1,024 Agents, Boosting Pass Rate 62%
donglixp · x · 2026-09-24
Microsoft Research introduces Agensh, a self-organized multi-agent harness with no central orchestrator that scales concurrent cooperation from 1 to 1,024 agents building large software systems from scratch.
The key idea is self-organization over central scheduling: workers asynchronously gather context, claim sub-tasks, share findings, verify, and merge progress through lightweight infrastructure—a shared workspace, a message interface, and shared context.
Key results:
- Building pandoc from scratch in 6h with no internet: test pass rate rises from 33.89% (1 agent) to 55.06% (1,024 agents), a +62% relative gain
- On the five hardest ProgramBench tasks (FFmpeg, gromacs, PHP-src, etc.), scaling 1→128 agents lifts mean pass rate from 19.31% to 28.78%
- As the organization grows, cooperation emergently evolves from peer coordination to multi-worker integration, standardized workflows, and specialized roles
The project page includes a full replay of the 1,024-agent collaboration.
Related event: Microsoft's Agensh Scales to 1,024 Self-Organized Agents(3 posts)→
More from coding & agent
- YC-backed Roma turns your to-do list into an agent command center across Gmail and Notion — testingcatalog · 2026-11-12
- Boubou: open-source CLI turns your laptop into a fully local voice AI hub — alexcovo_eth · 2026-11-12
- User automates entire video clipping workflow with Nous Research's Hermes agent — Teknium · 2026-11-12
- Tool-call failure rates jump 14x across providers, charts pinpoint inference stack updates — remilouf · 2026-11-12
- As models improve, writing good evals gets harder and more important — amankhan · 2026-11-12
- Prime agent ships with day-one multi-agent communication, runs in Minecraft unmodified — sarahcat21 · 2026-11-12