MetaGen: Self-Evolving Roles and Topologies for Multi-Agent LLM Reasoning

2026-01-27

AI Summary

Existing large language model–based multi-agent systems are constrained by fixed role libraries and static interaction topologies, limiting their ability to dynamically adapt to task requirements and resulting in high reasoning costs and insufficient flexibility. This work proposes a training-free, inference-time framework that dynamically generates task-specific role definitions, constructs a constraint-satisfying execution graph centered on a minimal backbone, and iteratively refines both role prompts and collaboration structures using lightweight feedback signals. For the first time, this approach enables joint dynamic evolution of the role space and collaboration topology without updating model weights. Experiments demonstrate that the method significantly outperforms strong baselines on code generation and multi-step reasoning tasks, achieving a superior trade-off between accuracy and reasoning efficiency.

Technology Category

• Multiagent Systems: Multiagent Learning
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Large language models are increasingly deployed as multi-agent systems, where specialized roles communicate and collaborate through structured interactions to solve complex tasks that often exceed the capacity of a single agent. However, most existing systems still rely on a fixed role library and an execution-frozen interaction topology, a rigid design choice that frequently leads to task mismatch, prevents timely adaptation when new evidence emerges during reasoning, and further inflates inference cost. We introduce MetaGen, a training-free framework that adapts both the role space and the collaboration topology at inference time, without updating base model weights. MetaGen generates and rewrites query-conditioned role specifications to maintain a controllable dynamic role pool, then instantiates a constrained execution graph around a minimal backbone. During execution, it iteratively updates role prompts and adjusts structural decisions using lightweight feedback signals. Experiments on code generation and multi-step reasoning benchmarks show that MetaGen improves the accuracy and cost tradeoff over strong multi-agent baselines.

Problem

• multi-agent LLM
• role adaptation
• interaction topology
• dynamic reasoning
• inference cost

Innovation

• dynamic role generation
• adaptive topology
• training-free framework
• multi-agent LLM reasoning
• inference-time adaptation