AI Agents Now Rewrite Their Own Skills Autonomously

Researchers have developed Memento-Skills, a framework that lets AI agents update and expand their own capabilities without retraining the underlying language model. It functions as an evolving external memory, allowing agents to improve progressively based on environmental feedback. The framework targets a key bottleneck in enterprise AI deployment, offering a practical alternative to costly fine-tuning or manual skill-building. Developers say it complements existing tools like OpenClaw and Claude Code by adding continual learning capabilities to autonomous agents running in production.