SentientAGI's EvoSkill v2 improves AI agent skills without changing the model, beating Microsoft's SkillOpt by 7.7 points

SentientAGI has released EvoSkill v2, a system that automatically improves AI agent skills without modifying the underlying model or runtime. It works by testing the agent, learning from its attempts, and rewriting its instructions to boost performance. According to the company's new paper, which compares the system against Microsoft Research's SkillOpt and Alibaba's SkillBoost on held-out tasks, EvoSkill v2 gains 23.6 points on τ3-bench banking with no skill using Claude Haiku 5.5, beats SkillOpt by 7.7 points on SpreadsheetBench v2, and by 5.4 points on τ3-bench banking.

✨ AI summary of the linked source — not editorial content.

Explore the topic

• Microsoft
• AI Agents