Microsoft Research open-sources Agensh, tests collaboration among 1,024 coding agentsMachine translation
Automatically verified and published · Generated and evidence-checked automatically; not reviewed by a human.
Agensh lets coding agents claim tasks and coordinate through a shared code repository without a central orchestrator. Microsoft Research tested up to 1,024 agents on ProgramBench; with GPT-5.6-sol, the mean test-pass rate across five tasks rose from 19.31% with one agent to 28.78% with 128 agents.
This source permits summary display only.
Read at the original source