Symbolic Guidance for LLM Agents in Distributed Multiagent Coordination

2026-05-25

Proceedings of the 25th International Conference on Autonomous Agents and Multiagent Systems

AI Summary

This study addresses the performance inconsistency of large language model (LLM) agents in distributed coordination caused by excessive reasoning autonomy, establishing "autonomy modulation" as a key design principle for the first time. Methodologically, this work proposes a symbol-guided taxonomy that constructs an autonomy spectrum ranging from natural language to algorithmic execution. Based on the AgentsNet benchmark, it integrates symbolic pseudocode guidance with LLM reasoning techniques. Experimental results demonstrate that intermediate autonomy levels significantly outperform both fully open-ended and entirely predefined paradigms. Furthermore, moderate symbolic guidance effectively enhances the consistency and overall performance of multi-agent collaboration in complex scenarios.

Abstract

Large language models (LLMs) are increasingly deployed as autonomous agents in multi-agent systems, yet their ability to reliably execute distributed coordination protocols remains poorly understood. While AgentsNet, a benchmark framework for distributed coordination among LLM agents, enables such coordination, granting full reasoning autonomy often leads to inconsistent or degraded performance in complex domains. We hypothesize that coordination can be improved by regulating agent autonomy through symbolic guidance derived from established algorithms. To investigate this, we introduce the Symbolic Guidance Taxonomy (SGT), which characterizes a spectrum of autonomy ranging from open-ended natural language reasoning to fully prescribed algorithmic execution, with intermediate levels providing partial pseudocode guidance. Our results show that intermediate autonomy levels consistently outperform both unguided agents and fully prescriptive specifications. These findings identify autonomy regulation as a key design principle for LLM-based distributed coordination.

Problem

• Large Language Models
• Multi-Agent Systems
• Distributed Coordination
• Agent Autonomy
• Symbolic Guidance

Innovation

• Symbolic Guidance
• Distributed Multiagent Coordination
• Autonomy Regulation
• Large Language Models
• Symbolic Guidance Taxonomy