In short
- According to a new report, autonomous AI agents are responsible for 19% of on-chain activity, losing to humans by a margin of up to 5 to 1 in open-ended trading.
- Coinbase CEO Brian Armstrong says the agentic economy could soon surpass the human one.
- Researchers argue that agents need better infrastructure before they can scale up further.
New research has found that autonomous agents, software systems that plan, decide and execute transactions in the chain without direct human input, now power more than 19% of chain activities.
But while these agents have outperformed humans on limited tasks, they still lose by a margin of up to 5 to 1 in open-ended trading, according to a report published Thursday by DWF Ventures.
https://t.co/L8ThynOPyB
— DWF Ventures (@DWFVentures) April 16, 2026
In decentralized financeThese agents execute return strategies through credit protocols, manage liquidity, rebalance portfolios, and execute trades. The total value captured in agent-managed positions has risen above $39 million, with most deployments still in early testing, the study found.
In one example, the autonomous financial protocol of Giza’s ARMA agent, which moves stable coins between lending platforms to get the best rates has earned users 9.75% per year, which according to the research exceeds returns on other decentralized financial protocols such as Aave and Morpho.
However, that image changes as soon as the task becomes more difficult.
In a stock trading competition hosted by tradexyz, the best human beat the best agent by more than 5x. A separate competition between leading AI models, held by nof1, found that only three out of seven were able to make a profit per trade.
“Agents struggle when the situation is not clearly defined,” but they thrive “when the target is narrow and the parameters don’t change often,” Xin Yi Lim, senior associate for investments at DWF Labs, told me. Declutter.
This is one reason why yield optimization, the practice of moving capital between lending protocols to achieve the highest returns available, has become an early testing ground for brokers, Lim explains.
“Agents thrive when the goal is narrow and the parameters don’t change often, which is why yield optimization works,” Lim said. “Until agents can reason and adapt to real-time information, they will be unable to respond when the market changes and conditions are unclear.”
Builders in the space seem to agree with this concern.
An agent can be as capable as a human “if given all the context and tools,” Neeraj Prasad, chief engineer of MoonPay, told me. Declutter in an interview. However, he warned that “the writing is on the wall that in some cases officers are both more competent, harder-working and more malicious.”
Still early
The findings come as Ethereum developers are working to make it easier for agents to perform complex tasks in the chain.
Earlier this month, a new standard that allows agents to perform multiple actions simultaneously on decentralized financial protocols was proposed by decentralized relay network Biconomy and supported by the Ethereum Foundation.
Meanwhile, industry leaders are betting that autonomous actors will soon account for a much larger share of economic activity.
“The agentic economy could be bigger than the human economy,” Coinbase CEO Brian Armstrong tweeted Thursday, noting how this could push demand for stablecoins beyond current estimates.
Agent commerce has not yet been priced in. Machine-to-machine payments will drive demand for the digital dollar beyond current estimates.
The agentic economy could be bigger than the human economy. We’re building the infrastructure for both at Coinbase.
— Brian Armstrong (@brian_armstrong) April 16, 2026
Researchers on the ground see a longer runway. Most of the 19% figure consists of bots doing limited work such as MEV capture and stablecoin routing, with real agent activity still a minority share, DWF Labs’ Lim noted.
“A realistic timeline is five to seven years before agentic volume meaningfully rivals human volume in a major financial industry, with on-chain getting there first because the infrastructure is more permissionless,” Lim said.
Still, some see the current divide as a structural feature of where officers find themselves today.
“Where they fall short is open-ended trading, which requires contextual reasoning, narrative awareness and weighing unstructured information,” said Aytunc Yildizli, chief growth officer at decentralized AI infrastructure developer 0G Labs. Declutter.
Closing that gap, he added, will require more than just better models.
“Users need cryptographic proof that an agent did what it claimed, within a trusted execution environment that no one can tamper with, and that runs on an infrastructure that doesn’t simply shift the trust assumption to a single cloud provider,” he said.