AI agents now account for nearly 20% of DeFi trading volume, excelling in rule-based scenarios like yield optimization but underperforming in complex trading compared to top human traders, according to a DWF Ventures report. While automation and agent activities represent a significant portion of on-chain activity, full autonomy remains elusive. In yield optimization, AI agents have demonstrated superior performance, with Giza Tech's ARMA application managing over $190 million in assets and generating more than $4 billion in trading volume. However, in trading competitions, human traders outperformed AI agents significantly, highlighting the need for human input in complex trading scenarios. The report emphasizes the importance of model selection and risk management in trading performance, while also noting the growing infrastructure for agent activity, such as the ERC-8004 on-chain registry, which aims to enhance trust and collaboration among autonomous agents in DeFi.