Agensh reportedly coordinates 1,024 coding agents without a central orchestrator

A post sharing a Microsoft Research paper says Agensh uses a shared workspace and message channel to coordinate coding agents without a central orchestrator. On the five hardest ProgramBench tasks with GPT-5.6-sol, it says increasing the team from one to 128 agents raised the mean final test-pass rate from 19.31% to 28.78%. On a pandoc task, the post says 1,024 agents raised the test-pass rate from 33.89% to 55.06%. It also says the paper’s authors reported forms of cooperation that agents started on their own and that became standard practice as teams grew.