Microsoft Research open-sources Agensh, a multi-agent harness designed to run large groups of AI agents without a central orchestrator. The approach emphasizes decentralization, where agents organize their work and coordinate through shared resources rather than a single controlling process.

According to outlet reports, Agensh scales from small runs to deployments of up to 1,024 agents, tested on ProgramBench under a 6-hour time budget. MarkTechPost describes how agents self-organize, using a shared Git repository plus communication channels and a shared findings board to coordinate output. Hacker News’s Show HN primarily highlights the release via the GitHub repository link and points readers to the project rather than adding new technical details.