Agensh scales multi-agent cooperation to 1,024 agents successfully
Agensh: Scaling Organizational Intelligence to 1,024 Agents
Computation and LanguageMultiagent Systems
Summary
Complex tasks can be done faster by having many agents work together. Existing methods rely on a central boss to manage tasks, which limits how many agents can join. The authors built Agensh, where agents organize themselves without a boss, sharing tasks and progress asynchronously. When tested, more agents led to noticeably better performance, with 1,024 agents solving tasks much more effectively than one alone. This suggests that increasing the number of cooperating agents is a promising way to improve general intelligence systems.
What this means in practice
- •For software development teams: Coordinate large numbers of AI agents to speed up complex programming benchmarks and tasks.
- •For distributed computing engineers: Design large-scale multi-agent architectures that do not rely on central control for task management.
Authors
Zhihao Zhan, Ting Song, Li Dong, Shaohan Huang, Jianxun Lian, Yan Xia, Furu Wei
Abstract
A multi-agent system can reduce latency on complex tasks by executing work concurrently. Several pioneering harness frameworks support multi-agent systems. However, the scalability of current multi-agent harnesses is often constrained by a central orchestrator's capacity to allocate tasks and coordinate workers. To address this limitation, we introduce Agensh, a scalable self-organized multi-agent harness without a central orchestrator: concurrent workers execute a multi-agent cooperation loop, continuously gathering context, claiming and self-assigning sub-tasks, taking action and sharing findings, verifying results, and merging progress in an asynchronous manner. The loop is supported by the agentic organization infrastructure comprising three components: a shared workspace holds proposed, ongoing, and completed work; a message interface lets workers communicate; and shared context retains reusable findings and work intentions. To test the scalability of Agensh, we evaluate it on the five hardest ProgramBench tasks with GPT-5.6-sol (high). Scaling from 1 to 128 agents raises the mean final test-pass rate from 19.31% to 28.78%, an approximately 49% relative improvement. Larger organizations reach comparable test-pass rates earlier. On pandoc, scaling from 1 to 1,024 agents raises the final test-pass rate from 33.89% to 55.06%. Worker trajectories further show that different forms of self-organized cooperation gradually emerges and standardizes as the organization grows. These results reveal the number of agents as a new scaling dimension for multi-agent organizations to expand the frontier of general intelligence, offering a practical solution for complex tasks under hard latency constraints or time budgets.