selfimproving-agent/Awesome-Self-Improving-Agents

What it solves

The field of self-improving AI agents is growing fast, but information is scattered across papers, blogs, talks, and code repositories. This project gathers all of that into a single, organized hub, making it easy to find and follow the latest developments without digging through dozens of sources.

How it works

The repository is built around a clear taxonomy that splits self-improving agents into two main kinds: those that improve the underlying model (via fine-tuning, giving feedback, or exploring new experiences) and those that improve the surrounding scaffolding, such as prompts, memory, tools, and workflows. For each category it lists relevant papers, links to code, and other materials. It also includes a survey paper that explains the core ideas, plus an animated comparison showing how an agent differs from a standalone model.

Who it’s for

This is for anyone interested in AI agents and self-improvement techniques—researchers, students, engineers, and hobbyists. If you want a structured overview of current work, a starting point for your own projects, or a place to share resources, this hub is set up for you.

Highlights

• A curated list of papers, benchmarks, blogs, podcasts, interviews, videos, and courses.
• A taxonomy that classifies self-improving agents into foundation model improvement and scaffolding improvement.
• Direct links to code for many papers, making it easier to reproduce results.
• An active Discord community and an open contribution model via pull requests.
• The accompanying survey paper provides a comprehensive guide to the field.