Towards Multi-Agent Reasoning Systems for Collaborative Expertise Delegation: An Exploratory Design Study

TL;DR

This study investigates how to design collaborative reasoning systems with multiple expert agents. It formalizes sequential multi-agent reasoning as $\mathcal{Y} = \mathcal{F}(\mathcal{A}_1(\mathcal{Q},\mathcal{S}), ..., \mathcal{A}_n(\mathcal{Q},\mathcal{S}))$ and examines Expertise-Domain Alignment, Collaboration Paradigm, and System Scale. Across four MMLU-pro domains (Math, Health, Business, Law) using a 3-agent setup with a sequential communication protocol, it finds that aligning agent expertise to the task domain most benefits contextual reasoning, while math-focused tasks derive smaller gains. It shows that Diversity-Driven Perspective Integration consistently outperforms Structured Workflow, with response diversity correlating with improved performance, and that scaling to 6–10 agents provides larger gains for non-mathematical tasks but increases token overhead, underscoring the need for efficient inter-agent communication. Together, these results offer concrete guidelines for configuring specialized multi-agent reasoning systems and pinpoint key bottlenecks in scalable collaboration.

Abstract

Designing effective collaboration structure for multi-agent LLM systems to enhance collective reasoning is crucial yet remains under-explored. In this paper, we systematically investigate how collaborative reasoning performance is affected by three key design dimensions: (1) Expertise-Domain Alignment, (2) Collaboration Paradigm (structured workflow vs. diversity-driven integration), and (3) System Scale. Our findings reveal that expertise alignment benefits are highly domain-contingent, proving most effective for contextual reasoning tasks. Furthermore, collaboration focused on integrating diverse knowledge consistently outperforms rigid task decomposition. Finally, we empirically explore the impact of scaling the multi-agent system with expertise specialization and study the computational trade off, highlighting the need for more efficient communication protocol design. This work provides concrete guidelines for configuring specialized multi-agent system and identifies critical architectural trade-offs and bottlenecks for scalable multi-agent reasoning. The code will be made available upon acceptance.

Figures (8)

• Figure 1: Workflow diagram for a multi-agent reasoning system with specialized agents.
• Figure 2: Demonstration of three key dimensions characterizing research on multi-agent collaborative reasoning systems.
• Figure 3: Heatmap illustrating the correlation between specialized group expertise and task domains. Deeper colors indicate stronger correlations.
• Figure 4: Comparative analysis of diversity-driven versus structured workflow collaboration paradigms. Positive values signify Diversity-Driven's advantage over Structured Workflow.
• Figure 5: Illustration of response diversity across four distinct domains, where lower inter-agent response similarity corresponds to higher diversity.