挖掘以愈合：通过可解释的动态决策路径扩展通用代理协作

The rising agentic AI paradigm leverages the capabilities of multiple general-purpose large language model (LLM) agents to collaboratively tackle complex tasks. While many agentic AI systems employ predefined workflows or agent roles to mitigate complexity, the ideal scenario involves truly autonomous agents capable of emergent collaboration as the number of collaborating agents increases. However, such unstructured interactions often lead to redundant efforts and cascading failures that are challenging to interpret or rectify. This research investigates multi-agent systems composed of general-purpose LLM agents operating without predefined roles, control flows, or communication constraints, relying instead on dynamic decision paths to facilitate effective collaboration. By incorporating explainability mechanisms, the efficiency and reliability of multi-agent systems are enhanced.