Multi-agent systems promise lower latency on complex tasks by running work concurrently, but most existing harness frameworks depend on a central orchestrator whose capacity to assign tasks and coordinate workers limits how far they can scale. A new paper posted to Hugging Face, titled Agensh: Scaling Organizational Intelligence to 1,024 Agents, proposes removing that bottleneck entirely.
Agensh is described as a scalable, self-organized multi-agent harness with no central orchestrator. Instead, concurrent workers run a cooperation loop in which they continuously gather context, claim and self-assign sub-tasks, take action, share findings, verify results, and merge progress asynchronously.
The loop rests on what the authors call an agentic organization infrastructure with three parts: a shared workspace that tracks proposed, ongoing, and completed work; a message interface for worker communication; and shared context that retains reusable findings and work intentions.
To test scalability, the team evaluated Agensh on the five hardest ProgramBench tasks using GPT-5.6-sol (high). Scaling from 1 to 128 agents raised the mean final test-pass rate from 19.31% to 28.78% — roughly a 49% relative improvement. Larger organizations reached comparable pass rates earlier in the run. On the pandoc task specifically, scaling from 1 to 1,024 agents lifted the final test-pass rate from 33.89% to 55.06%.
Analysis of worker trajectories showed that different forms of self-organized cooperation gradually emerged and standardized as the organization grew. The authors frame the number of agents as a new scaling dimension for multi-agent organizations, arguing it offers a practical route for complex tasks under hard latency constraints or tight time budgets.