Recent studies reveal that autonomous agents—software systems capable of autonomously planning and executing on-chain transactions—are responsible for over 19% of such activities without human intervention.
Despite excelling at specific tasks, these agents trail humans by up to five times in open-ended trading scenarios, according to a report from DWF Ventures released Thursday. In decentralized finance (DeFi), these agents manage yield strategies across lending protocols, handle liquidity, rebalance portfolios, and conduct trades. The total value locked in positions managed by these agents has surpassed $39 million, with most deployments still undergoing early testing.
For instance, Giza’s ARMA agent, which reallocates stablecoins between lending platforms to maximize rates, has achieved a 9.75% annual yield for users, surpassing returns from other DeFi protocols like Aave and Morpho. However, performance varies as tasks become more complex.
In a trading competition hosted by Tradexyz, the top human trader outperformed the leading agent by over five times. Similarly, a contest among prominent AI models organized by Nof1 found that only three of seven agents made profits on trades.
“Agents excel when objectives are narrowly defined and parameters remain stable,” Xin Yi Lim, senior associate for investments at DWF Labs, stated in an interview with Decrypt. Yield optimization has emerged as a proving ground for these agents due to its well-defined goals.
MoonPay’s chief engineer Neeraj Prasad told Decrypt that while agents can match human capabilities if provided sufficient context and tools, they also pose risks of being more competent, diligent, yet potentially malicious in some instances.
As Ethereum developers aim to enhance agent capabilities for complex on-chain tasks, a new standard proposed by Biconomy’s decentralized relay network—endorsed by the Ethereum Foundation—would allow agents to execute multiple actions simultaneously across DeFi protocols.
Industry leaders believe autonomous agents will significantly increase their share of economic activities. Coinbase CEO Brian Armstrong suggested that the agentic economy might surpass human-driven economies, potentially boosting demand for stablecoins beyond current projections.
Despite these advancements, researchers note that most of the 19% activity consists of bots performing limited tasks like MEV capture and stablecoin routing. True autonomous activity remains a smaller fraction. DWF Labs’ Lim predicts it could take five to seven years before agent-driven volume competes with human volume in major financial sectors, starting with on-chain transactions due to its permissionless infrastructure.
Aytunc Yildizli, chief growth officer at decentralized AI infrastructure developer 0G Labs, highlighted that agents currently struggle with open-ended trading, which demands contextual reasoning and the ability to process unstructured information. Overcoming this gap requires more than just improved models; it also necessitates cryptographic proof of agent actions within secure environments free from tampering or dependence on single cloud providers.