GBC: Gradient-Based Connections for Optimizing Multi-Agent Systems

2026-06-26

AI Summary

This work addresses the limitations of multi-agent systems stemming from insufficient coordination and coarse-grained credit assignment. The authors propose a gradient-driven, fine-grained attribution mechanism that models multi-agent collaboration as a computational graph, leveraging backpropagated task loss signals at the token level to quantify each agent’s contribution to downstream performance. This approach constructs an attribution graph to precisely localize error sources and refine prompts. For the first time, it enables accurate credit assignment across agents and reasoning steps by integrating large language model architectures with prefix gradient computation (AgentChord). Experiments on MultiWOZ and τ-bench demonstrate substantial improvements over existing single- and multi-agent baselines, confirming a strong positive correlation between high-quality attribution and overall system performance.

Abstract

Multi-agent systems (MAS) built on large language models (LLMs) provide a promising framework for solving complex tasks through role specialization and structured interaction. However, their performance is often limited by miscoordination and, more fundamentally, the lack of fine-grained credit assignment across agents. Existing approaches typically rely on coarse-grained feedback, making it difficult to identify which agents or interaction steps are responsible for errors. We propose Gradient-Based Connections (GBC), an approach for fine-grained attribution and optimization of multi-agent systems. GBC models a MAS as a computational graph and introduces gradient-based connection weights to quantify the influence of each agent's output on downstream agents at the token level. By constructing an attribution graph and propagating task-specific loss signals backward, our method enables precise identification of error sources and targeted prompt optimization. We further develop AgentChord, an efficient implementation that leverages prefix-based gradient computation. Experiments on MultiWOZ and τ-bench show that GBC improves multi-agent performance and outperforms strong single-agent and multi-agent baselines, and higher attribution quality is associated with greater optimization effectiveness. Code is available at: https://github.com/yxc-cyber/AgentChord.

Problem

• multi-agent systems
• credit assignment
• fine-grained attribution
• coordination
• large language models

Innovation

• Gradient-Based Connections
• Multi-Agent Systems
• Fine-Grained Attribution
• Computational Graph
• Prompt Optimization