AgentFactory: A Self-Evolving Framework Through Executable Subagent Accumulation and Reuse

TL;DR

AgentFactory introduces a novel framework where executable subagents are accumulated and refined through self-evolution, enabling continuous improvement and efficiency in complex task execution using Python code.

Contribution

It proposes a new self-evolution paradigm that preserves successful solutions as executable code, enhancing robustness and portability over traditional textual prompt methods.

Findings

Subagents are continuously refined and accumulated over time.

The framework reduces effort for similar tasks without manual intervention.

Open-source implementation demonstrates practical effectiveness.

Abstract

Building LLM-based agents has become increasingly important. Recent works on LLM-based agent self-evolution primarily record successful experiences as textual prompts or reflections, which cannot reliably guarantee efficient task re-execution in complex scenarios. We propose AgentFactory, a new self-evolution paradigm that preserves successful task solutions as executable subagent code rather than textual experience. Crucially, these subagents are continuously refined based on execution feedback, becoming increasingly robust and efficient as more tasks are encountered. Saved subagents are pure Python code with standardized documentation, enabling portability across any Python-capable system. We demonstrate that AgentFactory enables continuous capability accumulation: its library of executable subagents grows and improves over time, progressively reducing the effort required for similar tasks without manual intervention. Our implementation is open-sourced at https://github.com/zzatpku/AgentFactory, and our demonstration video is available at https://youtu.be/iKSsuAXJHW0.