AgentAsk: Multi-Agent Systems Need to Ask

2025-10-08

AI Summary

Multi-agent LLM systems often underperform single-agent baselines due to cascading edge errors in message passing. To address this, we propose AgentAsk—a lightweight, plug-and-play clarification module that interrupts error propagation by inserting minimal, contextually necessary queries into interaction chains. We introduce a fine-grained edge-error taxonomy, design a link-local intervention mechanism, and formulate a three-stage decision optimization framework governing *when*, *what*, *whom*, and *how* to query. A compact policy model—trained via failure-trajectory distillation and optimized online with an E-GRPO reinforcement learning objective—enables architecture-agnostic, zero-configuration deployment. Evaluated on mathematical reasoning, logical deduction, and code-generation benchmarks, AgentAsk achieves significant gains in accuracy and robustness, with marginal overhead: latency and computational cost increase by less than 5%, while performance approaches that of strong oracle evaluators.

Technology Category

• Multiagent Systems: Agent Communication
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Search and Optimization: Learning to Search

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
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Multi-agent systems built on large language models (LLMs) promise enhanced problem-solving capabilities through collaborative division of labor. However, they frequently underperform single-agent baselines due to edge-level error cascades: minor inaccuracies at one message handoff propagate across the entire chain. We propose AgentAsk, a lightweight and plug-and-play clarification module that treats every inter-agent message as a potential failure point and inserts minimally necessary questions to arrest error propagation. AgentAsk follows a three-stage pipeline: (i) distilling edge-level judgments from curated failure traces into a compact policy, (ii) supervising the policy to determine when/what/whom/how to ask, and (iii) optimizing online with E-GRPO, a reinforcement learning objective that balances accuracy, latency, and cost. The module is architecture-agnostic and easy to integrate into existing orchestration. Across math, reasoning, and coding benchmarks, AgentAsk consistently improves accuracy and robustness over public multi-agent implementations while keeping overhead minimal, with latency and extra cost all less than 5%, approaching the performance of a strong evaluator. Beyond empirical improvements, we contribute a principled taxonomy of edge-level errors and a practical recipe for link-local intervention, offering a scalable pathway toward more reliable LLM-based multi-agent systems.

Problem

• Preventing error propagation in multi-agent LLM systems
• Reducing cascade failures through targeted clarification questions
• Balancing accuracy with minimal latency and cost overhead

Innovation

• Plug-and-play clarification module prevents error propagation
• Three-stage pipeline distills policies and supervises questioning
• E-GRPO optimization balances accuracy with latency and cost