This repository is a curated list of research papers and resources on self-evolving AI agents, covering advancements from 2023 to 2025. It categorizes techniques into single-agent optimization, multi-agent optimization, and domain-specific optimization, providing a valuable overview for researchers and practitioners in the field of artificial intelligence.

How It Works

The project presents a taxonomy of AI agent evolution and optimization techniques. It organizes approaches based on their development trajectory, highlighting methods for optimizing LLM behavior, prompts, memory, and tools within single-agent frameworks. It also covers multi-agent optimization strategies and domain-specific advancements, offering a structured view of the evolving landscape.

Quick Start & Requirements

This is a curated list of research papers and does not have a direct installation or execution command. The primary requirement is an interest in the field of AI agents and their evolution. Links to papers and code are provided for further exploration.

Highlighted Details

• Comprehensive coverage of single-agent optimization techniques including LLM behavior, prompt, memory, and tool optimization.
• Extensive categorization of multi-agent optimization strategies and domain-specific advancements.
• Includes papers on evaluation methods like LLM-as-a-Judge and Agent-as-a-Judge, as well as safety and alignment research.
• Features a visual taxonomy of AI agent evolution and optimization techniques.

Maintenance & Community

The project is marked as "still cooking" with a welcome for contributions. Contact information via email is provided for questions and suggestions.

Licensing & Compatibility

The repository is licensed under the MIT License.

Limitations & Caveats

The project is a survey and is still under development, as indicated by the "We're still cooking — Stay tuned!" message. Some papers are marked with "Arxiv" or specific conference years, suggesting ongoing research and potential for future updates.