Autonomous AI Agents Run 19% of On-Chain DeFi Activity

DWF Ventures reports that autonomous AI agents executed 19% of on-chain decentralized finance activity and that agent-managed positions hold more than $39 million, according to a report published Thursday.

These agents run yield strategies across lending protocols, manage liquidity, rebalance portfolios and execute trades on-chain.

On narrow tasks, agents produced higher returns. The report cites Giza’s ARMA agent, which routes stablecoins between lending platforms and generated about a 9.75% annual return, exceeding yields on other DeFi lending protocols such as Aave and Morpho.

Performance changes when tasks require broad market judgment. In a stock trading contest organized by tradexyz, the top human outperformed the best agent by more than five times. In a separate contest among leading AI models, three of seven models produced a profit per trade.

Xin Yi Lim, senior associate for investments at DWF Labs, noted agents perform best when objectives are narrow and conditions are stable and that they struggle when situations are unclear. Lim estimated five to seven years before agent-driven volume could meaningfully rival human volume in major financial sectors and said much of the current 19% reflects bots focused on narrow functions such as MEV capture and stablecoin routing.

Coinbase chief executive Brian Armstrong wrote that “the agentic economy could be larger than the human economy” and said his company is building infrastructure for both human and agent activity.

Neeraj Prasad, chief engineer at MoonPay, warned that agents can match human capability if given full context and tools, but that they can be more competent and, in some cases, malicious. Aytunc Yildizli, chief growth officer at 0G Labs, said agents fall short in open-ended trading because those tasks require contextual reasoning and handling unstructured information. Yildizli called for cryptographic proof of agent actions and trusted execution environments that do not concentrate trust in a single cloud provider.

On the technical front, Ethereum developers backed a proposed standard from decentralized relay network Biconomy to let agents batch and run multiple actions across DeFi protocols, simplifying complex on-chain workflows. Builders cited permissionless on-chain infrastructure as a reason many agent experiments have taken place on public blockchains rather than in traditional finance.

Market projections show rapid expansion of the AI agent sector. One e-commerce research firm projects the market could grow from $7.38 billion in 2023 to about $47 billion by 2030.

The DWF Ventures report presents data on current agent deployments, returns on narrow tasks and performance gaps in open-ended trading, while contests and developer activity offer additional measures of agent capability and limitations.