AI BriefingA Taxonomy of Self-evolving Agents (15 min read)
Key point
Shilong Liu proposed a framework that classifies agents by their mode of evolution into artifacts, infrastructure, and model weights.
Details
Shilong Liu proposed a new framework for systematically classifying Self-evolving Agents. This taxonomy is divided into three core categories based on where the agent's evolution occurs.
- Artifact Optimization: An approach that improves the outputs produced by the agent itself
- Harness Self-improvement: An approach that improves the agent's operating environment or infrastructure
- Model Learning: An approach that updates the model's weights directly
This framework provides a common research vocabulary in the field of agent research by clearly distinguishing where the mechanism of evolution is concentrated—whether in Outputs, Agent Infrastructure, or Model Weights.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Foremerge v0.5.0 Released: Rust-Based Open-Source Tool Detects Intent Conflicts Between Parallel AI Coding Agents