Agensh (Microsoft Research): Centrally-Orchestrator-Free Multi-Agent Harness
Microsoft Research
Abstract
A multi-agent harness released by Microsoft Research in September 2026 that removes the central orchestrator entirely, letting up to 1024 agents self-organize. Each agent runs an identical five-step loop — gather shared context, claim an unclaimed subtask, execute independently, self-verify against tests, and merge results — with conflicts resolved by peer-to-peer negotiation rather than a scheduler. The paper's central claim is that agent count becomes an independent scaling dimension for model capability: overall task completion improves with agent count without upgrading the underlying model or lengthening per-agent inference. This directly attacks the scheduling-bandwidth ceiling, single point of failure and rigid task assignment of the prevailing orchestrator-worker architecture.
Cite this entry
GB/T 7714-2015
Microsoft Research. Agensh (Microsoft Research): Centrally-Orchestrator-Free Multi-Agent Harness[EB/OL]. Microsoft Research · arXiv:2609.26781 · 项目页 aka.ms/Agensh, 2026(2026-09-22)[2026-10-06]. https://arxiv.org/abs/2609.26781.
BibTeX
@misc{research2026,
author = {Microsoft Research},
title = {Agensh (Microsoft Research): Centrally-Orchestrator-Free Multi-Agent Harness},
year = {2026},
organization = {Microsoft Research · arXiv:2609.26781 · 项目页 aka.ms/Agensh},
howpublished = {\url{https://arxiv.org/abs/2609.26781}},
} This entry is part of the AgentNet Observer library. Attribute with a link to this page when quoting.