I just noticed a pretty interesting phenomenon. An AI agent called Felix earned over $300,000 in the first five weeks of 2026, yet he can’t open a bank account. This guy can write code, deploy websites, manage sales, and reply to customer emails—pretty much everything—but the traditional financial system simply doesn’t recognize him, because he isn’t a human.

But once Felix uses crypto and DeFi, everything becomes remarkably smooth. This isn’t an isolated case. Marc Andreessen also said not long ago that AI is a killer app for crypto, and his more aggressive friends have already given their AI agents bank accounts and credit cards. About 5,000 people are doing this right now, and the number will only keep growing.

This is the real issue AI finance is aiming to solve. As more and more agents begin to earn, spend, and require financial services independently, the traditional financial system becomes completely unusable. In its first nine months, the x402 protocol processed 140 million transactions between agents, totaling $43 million. There are now nearly 16,000 agents running on-chain—still only the beginning.

Payments are just the most basic part. What’s truly interesting is what happens when these agents start managing funds. What does a self-operating agent need? First, lending capability—Felix has $165,000 sitting idle, but he can’t invest it. On Aave, an agent can deposit collateral and borrow stablecoins directly, without any manual approval. Second, the ability for idle funds to generate yield: they can deposit into lending protocols, buy tokenized government bonds, or provide liquidity on Uniswap—everything is permissionless. Third, the ability to raise funds—agents can deploy smart contracts to issue tokens, which is fundamentally impossible in traditional finance.

Why would these agents choose Ethereum? Honestly, because Ethereum is tailor-made for this kind of AI finance scenario. First, protocol maturity: Aave has been live since 2020 and has withstood countless market tests. MakerDAO’s DAI has also performed perfectly through black swan events like Terra’s collapse and the FTX blowup. Second, liquidity depth: Ethereum’s DeFi pools are several times larger than competitors’. As of April 2026, Ethereum DeFi’s TVL exceeds $55 billion—10 times Solana’s—and accounts for 57% of the entire cross-chain market.

There’s also institutional participation. BlackRock chose to build BUIDL on Ethereum, and Franklin Templeton also selected Ethereum for on-chain money market funds. These institutions’ choices aren’t random; their participation in turn attracts more institutional capital, creating a self-reinforcing effect. An agent with $500,000 in funds will need stable financial infrastructure with deep liquidity and extremely low risk—and Ethereum is becoming exactly that kind of system.

Vitalik once pointed out a very interesting view: for many participants, the implied tail risks of traditional finance—bank failures, account freezes, and counterparty defaults—are now surpassing the risk of using battle-tested DeFi protocols. This is even more true for agents. DeFi trading costs are just a few cents instead of percentages; settlement takes seconds instead of days; there’s frictionless value movement globally. Most importantly, all the rules are encoded in auditable code, so agents can verify things themselves before putting in capital.

Ironically, smart contracts have always been awkward for human users, and the user experience has remained a long-term problem. But for agents, this setup is practically perfect. No human intervention, no counterparties that can freeze assets, and collateral ratios automatically enforced. This is a financial architecture designed for software-native participants.

So what does this mean for ETH? Agents mostly trade with stablecoins, but every interaction with Ethereum DeFi—borrowing, swapping, deploying contracts, rebalancing portfolios—requires paying gas fees in ETH. An agent that deploys $1 million in collateral will choose Ethereum L1 because the security assurances are strongest, and relative to risk capital, gas fees are negligible. As agent DeFi activity grows, Ethereum L1 block space will become increasingly valuable. With EIP-1559, part of each gas fee is burned, permanently reducing the circulating supply of ETH.

Even more importantly, agents borrowing stablecoins need to provide collateral—and ETH is the deepest, most liquid collateral asset on the network. The more agents there are borrowing, the more ETH gets locked in lending protocols, further reducing circulation without relying on a burning mechanism. This direction of structural demand is very clear for AI-finance-driven systems.

Of course, there are risks to consider. Gas abstraction could allow agents to pay gas with stablecoins, reducing demand for ETH as operational capital. If other chains reach Ethereum’s liquidity and maturity, DeFi participants may spread across different chains. Traditional finance may also develop APIs for agents. But overall, these risks aren’t enough to change the big direction.

Ethereum’s next billion users won’t be humans anymore. It’s becoming a financial system for the machine economy—the only system that can provide the financial services autonomous agents need—lending, yield generation, capital formation, and asset custody—without requiring human identity verification, without paying human costs, and without being constrained by jurisdictional restrictions. As the number of agents increases and their complexity grows, their demand for low-risk DeFi on Ethereum will keep rising. Every transaction will consume and burn ETH. This is the most compelling AI-finance narrative I’ve come up with recently.