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

2026-03-18 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, quick

Summary

AgentFactory introduces a novel self-evolution framework for LLM-based agents that preserves successful task solutions as executable Python subagent code, rather than relying on textual prompts or reflections. This approach allows subagents to be continuously refined through execution feedback, enhancing their robustness and efficiency with each new task. The framework enables continuous capability accumulation, where the library of executable subagents grows and improves over time, progressively reducing the effort for similar tasks without requiring manual intervention. These subagents are pure Python code with standardized documentation, ensuring portability across various Python-capable systems. The implementation of AgentFactory is open-sourced.

Key takeaway

For AI Architects and Research Scientists developing LLM-based agents, AgentFactory offers a robust paradigm for continuous capability accumulation. By adopting its method of preserving successful task solutions as executable Python subagent code, you can significantly enhance agent reliability and efficiency over time. This approach reduces manual intervention and ensures portability, accelerating development cycles for complex, evolving agent systems.

Key insights

AgentFactory uses executable Python subagent code for LLM agent self-evolution, improving reliability and efficiency.

Principles

• Preserve solutions as executable code.
• Refine subagents via execution feedback.
• Standardize code for portability.

Method

AgentFactory preserves successful task solutions as executable Python subagent code, which is then continuously refined based on execution feedback to improve robustness and efficiency.

In practice

• Implement self-evolving agents with Python.
• Automate task re-execution in complex scenarios.
• Reduce manual intervention for repetitive tasks.

Topics

• LLM Agents
• Self-Evolving Systems
• Executable Code Generation
• Subagent Accumulation
• Task Automation

Code references

• zzatpku/AgentFactory

Best for: AI Architect, AI Scientist, Research Scientist, AI Researcher, AI Engineer, Machine Learning Engineer

Related on AIssential

• Programming with LLM Agents in 2025 — LLM agents, like Open Hands, significantly amplify developer productivity by automating iterative coding and visualization tasks.
• Beyond Individual Intelligence: Surveying Collaboration, Failure Attribution, and Self-Evolution in LLM-based Multi-Agent Systems — Multi-agent systems need integrated frameworks for collaboration, fault attribution, and self-evolution to achieve collective intelligence.
• New review paper argues code is how AI agents think and act, not just what they produce — Code forms the executable, traceable, and stateful foundation for AI agent reasoning and action.
• Single Agent vs Multi-Agent: When to Build a Multi-Agent System — AI agents leverage LLMs, tools, and memory, often using the ReAct pattern, to perform complex tasks through reasoning and action.
• Claude Code + Karpathy's NEW Self-Evolving System = 10x Code Generation — AI agents can build and maintain self-evolving knowledge systems using structured wikis and schema rules.
• SEAGym: An Evaluation Environment for Self-Evolving LLM Agents — Self-evolving LLM agent evaluation requires comprehensive metrics beyond isolated scores to understand harness updates.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.