Microsoft Research Introduces SkillOpt: Turning Agent Skills into Trainable Parameters

Jul 10, 2026

Microsoft Research has introduced SkillOpt, a method that converts agent skill editing into a training process, aiming to improve the reliability of AI agents without changing model weights. This approach addresses the issue of AI agents failing due to manual skill modifications that do not guarantee better performance.

Why it matters: SkillOpt could make AI agents more reliable by treating skills as trainable parameters, reducing reliance on manual tuning.

Full story at: Microsoft Research ↗