MAS$^2$: Self-Generative, Self-Configuring, Self-Rectifying Multi-Agent Systems

2025-09-29

AI Summary

Conventional multi-agent systems (MAS) suffer from poor adaptability to dynamic environments due to static agent deployment. Method: This paper proposes a “Generate-Execute-Correct” tri-agent collaborative framework to construct a dynamic MAS endowed with recursive self-generation, automatic configuration, and online correction capabilities. A collaboration tree is employed to optimize meta-agent training, while large language models (LLMs) enable automated role design, tool invocation, and communication protocol evolution—supporting cross-model generalization and cost-performance trade-off optimization. Contributions/Results: Evaluated on seven benchmark tasks, the framework achieves an average performance improvement of 19.6%, a maximum cross-LLM generalization gain of 15.1%, and significantly improves the cost-performance Pareto frontier.

Technology Category

• Multiagent Systems: Multiagent Learning
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The past two years have witnessed the meteoric rise of Large Language Model (LLM)-powered multi-agent systems (MAS), which harness collective intelligence and exhibit a remarkable trajectory toward self-evolution. This paradigm has rapidly progressed from manually engineered systems that require bespoke configuration of prompts, tools, roles, and communication protocols toward frameworks capable of automated orchestration. Yet, dominant automatic multi-agent systems, whether generated by external modules or a single LLM agent, largely adhere to a rigid `` extit{generate-once-and-deploy}'' paradigm, rendering the resulting systems brittle and ill-prepared for the dynamism and uncertainty of real-world environments. To transcend this limitation, we introduce MAS$^2$, a paradigm predicated on the principle of recursive self-generation: a multi-agent system that autonomously architects bespoke multi-agent systems for diverse problems. Technically, we devise a `` extit{generator-implementer-rectifier}'' tri-agent team capable of dynamically composing and adaptively rectifying a target agent system in response to real-time task demands. Collaborative Tree Optimization is proposed to train and specialize these meta-agents. Extensive evaluation across seven benchmarks reveals that MAS$^2$ achieves performance gains of up to $19.6%$ over state-of-the-art MAS in complex scenarios such as deep research and code generation. Moreover, MAS$^2$ exhibits superior cross-backbone generalization, effectively leveraging previously unseen LLMs to yield improvements of up to $15.1%$. Crucially, these gains are attained without incurring excessive token costs, as MAS$^2$ consistently resides on the Pareto frontier of cost-performance trade-offs. The source codes are available at https://github.com/yeyeyeah2/MAS2.

Problem

• Overcoming rigid generate-once-and-deploy multi-agent system limitations
• Enabling autonomous dynamic adaptation to real-time task demands
• Achieving robust performance in complex dynamic environments

Innovation

• Autonomously generates multi-agent systems for diverse problems
• Dynamically composes and adaptively rectifies agent systems
• Uses generator-implementer-rectifier tri-agent team architecture