From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents

2026-07-08

AI Summary

This work addresses the limitations of current large language model agents, which rely on static sets of atomic action tools and must repeatedly reconstruct low-level reasoning for recurring tasks—leading to high computational overhead and increased failure rates. To overcome this, the authors propose EvoSOP, a novel framework that enables agent self-evolution through iterative refinement of Standard Operating Procedures (SOPs). EvoSOP automatically extracts SOPs from execution trajectories, constructs higher-order reusable tools, and incorporates a multi-stage tool lifecycle management mechanism encompassing construction, merging, evaluation, and pruning. This approach substantially improves success rates on complex tasks, significantly reduces the number of interaction rounds, and enhances the agent’s capacity for abstraction and tool reuse, offering a scalable pathway toward self-evolving intelligent agents.

Abstract

Tool utilization enables Large Language Model (LLM) agents to interact with the real world and resolve complex tasks. However, existing agent frameworks predominantly rely on static toolsets composed of granular atomic actions (e.g., basic file I/O or single-turn search), which forces agents to reinvent low-level logic for every recurring workflow, leading to increased reasoning overhead and failure rates. In this study, we propose that agents can achieve self-evolution by synthesizing these atomic actions into reusable Standard Operating Procedures (SOPs), which function as callable higher-order tools that encapsulate multi-step logic. We further introduce EvoSOP, a framework that empowers agents to extract SOPs from execution trajectories and iteratively optimize the toolset through a systematic lifecycle of construction, merging, evaluation, and pruning. Extensive experiments demonstrate that EvoSOP significantly boosts task success rates while substantially reducing the number of interaction rounds compared to baselines. Our analysis also reveals that iterative tool optimization fosters reliable and efficient tool-use patterns, providing a scalable pathway for the development of self-evolving agents.

Problem

• tool utilization
• atomic actions
• Standard Operating Procedures
• self-evolving agents
• reasoning overhead

Innovation

• Standard Operating Procedures
• Self-Evolving Agents
• Tool Optimization
• Large Language Models
• Iterative Learning