RA-MoWE: Workflow-Affinity Embeddings for Query Clustering and Agentic Workflow Generation

2026-10-06

AI Summary

This study addresses the efficiency trade-off between universal optimization and per-query search in agent workflows by proposing an expert workflow reuse framework based on workflow-affinity embeddings. The method introduces a novel embedding paradigm constructed from the performance of fixed reference workflows, clustering queries to generate reusable expert workflows that are iteratively refined through execution feedback and large language model coordination. Furthermore, it enables zero-execution-cost matching for new queries via encoder-based prediction. Experimental results demonstrate that the proposed approach improves the average task score by 4.04 percentage points across 300 test queries while reducing the number of language model calls by 27.7%.

Abstract

Agentic workflows enable large language models (LLMs) to solve complex tasks by coordinating reasoning, tool use, and verification. However, a workflow optimized for an entire task collection can overlook differences in the reasoning strategies that individual queries need, while searching for a new workflow for every query repeats costly optimization. To address this tradeoff, we introduce RA-MoWE, a framework that uses workflow-affinity embeddings to cluster queries and guide the generation of reusable expert workflows. Each embedding records how well a fixed set of reference workflows solves a query, revealing similarities in which reasoning strategies are effective. RA-MoWE uses each cluster's queries and average embedding to initialize and refine a specialized workflow through execution feedback. An embedding encoder predicts these embeddings from query text, allowing new queries to select a generated expert without first executing the reference workflows. On a 300-query test set drawn from four benchmarks spanning mathematics, science, and programming, RA-MoWE improves average task score by 4.04 percentage points over selecting among the reference workflows, while using 27.7% fewer language-model calls at inference.

Problem

• Agentic Workflows
• Query Clustering
• Large Language Models
• Workflow Optimization

Innovation

• Workflow-Affinity Embeddings
• Query Clustering
• Agentic Workflow Generation
• Embedding Encoder
• Execution Feedback