Google's WikiSkill boosts AI agent performance with persistent memory

google researchers have introduced wikiskill, a framework that gives ai agents a persistent knowledge base to improve performance over time. wikiskill collects knowledge about past failures and successes in a wiki-like structure and uses it to improve the agent's abilities. the framework consists of three layers: the raw layer, wiki layer, and skill layer. wikiskill consistently outperforms other skill evolution methods, boosting performance on five benchmarks, including math reasoning, web search, and document question-answering. larger models benefit more from wikiskill, but smaller models also show significant improvements.