ARM: Discovering Agentic Reasoning Modules for Generalizable Multi-Agent Systems

2025-10-07

AI Summary

Existing LLM-driven multi-agent systems (MAS) suffer from poor generalization across architectures, high computational overhead, and reliance on manually annotated validation sets. Method: We propose the Agentic Reasoning Module (ARM), a novel paradigm that replaces conventional chain-of-thought (CoT) reasoning with modular, evolvable, and recursively invocable reasoning units—enabling strong cross-model and cross-task generalization. ARM automatically discovers efficient reasoning modules via execution-trace-guided tree search and mutation in code space, and dynamically orchestrates them using a meta-controller. Contribution/Results: Experiments demonstrate that ARM significantly outperforms both human-designed and state-of-the-art automated MAS approaches—without task-specific tuning—across diverse foundation models and task domains. It is the first framework to achieve end-to-end evolvability, high generalization, and minimal human dependency in multi-agent reasoning architecture design.

Technology Category

• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Large Language Model (LLM)-powered Multi-agent systems (MAS) have achieved state-of-the-art results on various complex reasoning tasks. Recent works have proposed techniques to automate the design of MASes, eliminating the need for manual engineering. However, these techniques perform poorly, often achieving similar or inferior performance to simple baselines. Furthermore, they require computationally expensive re-discovery of architectures for each new task domain and expensive data annotation on domains without existing labeled validation sets. A critical insight is that simple Chain of Thought (CoT) reasoning often performs competitively with these complex systems, suggesting that the fundamental reasoning unit of MASes, CoT, warrants further investigation. To this end, we present a new paradigm for automatic MAS design that pivots the focus to optimizing CoT reasoning. We introduce the Agentic Reasoning Module (ARM), an agentic generalization of CoT where each granular reasoning step is executed by a specialized reasoning module. This module is discovered through a tree search over the code space, starting from a simple CoT module and evolved using mutations informed by reflection on execution traces. The resulting ARM acts as a versatile reasoning building block which can be utilized as a direct recursive loop or as a subroutine in a learned meta-orchestrator. Our approach significantly outperforms both manually designed MASes and state-of-the-art automatic MAS design methods. Crucially, MASes built with ARM exhibit superb generalization, maintaining high performance across different foundation models and task domains without further optimization.

Problem

• Automating multi-agent system design for complex reasoning tasks
• Eliminating computationally expensive architecture rediscovery per domain
• Enhancing generalization across models and tasks without re-optimization

Innovation

• Optimizes Chain of Thought reasoning via specialized modules
• Discovers modules through tree search and mutation
• Enables generalization across models and tasks