Ethereum (ETH) Registered AI Agents Double to 72,028 in Five Months as Joseph Lubin Bets on Convergence

ERC-8004 Registrations Double in Five Months

If Joseph Lubin is right, the next wave of artificial intelligence will not run inside closed corporate systems but on a shared financial platform where autonomous software commissions work, executes it and settles its own invoices, and Ethereum (ETH) would be the venue where that commerce lands. The Ethereum co-founder set out the argument in a CNBC interview on October 7: AI can supply creativity and expertise, he said, but it cannot manufacture trust itself. Blockchain, in his view, fills exactly that gap, and the two industries would converge into a single platform covering everything from task assignment and execution to the payment of rewards. The on-chain record already shows the identity layer being built. Under the smart contract standard ERC-8004, which tracks the identity and reputation of AI agents, Ethereum hosted 72,028 registered agents as of October 20, roughly 9.2% of the 783,881 agents registered across all networks. That count has more than doubled from 32,343 on May 13, a five-month run that outpaces any comparable agent economy on rival altcoin networks. Analysts who mapped the agent economy over the weekend frame the thesis plainly: an agent that needs to buy a service, move funds or prove who it is wants a venue with deep liquidity and a trust layer, and the Ethereum ecosystem currently offers both. What the numbers do not yet show is revenue. Registered agents pay nothing on their own; paying agents do. Whether the doubling of registrations converts into transaction volume is the question that matters for the Ethereum price, and the payment data so far argues for patience rather than momentum.

Liquidity Pulls, Payments Lag

The strongest card Ethereum (ETH) holds is money that is already there. Decentralized finance applications built on the network hold roughly $49.6 billion in deposits, and stablecoins circulating across it exceed $144 billion, the deepest machine-payable pool in crypto. An autonomous agent that wants to purchase a product or shift funds without asking permission needs exactly this: enough liquidity to transact in size, and payment rails that never close. That infrastructure is what pulls developers, and their software, onto the chain in the first place. The revenue line tells a slower story. Monthly payments routed through x402, a stablecoin settlement standard designed for machine-to-machine commerce, peaked at $20.5 million in November 2025 and had fallen to $263,863 by September 2026. Ethereum's network fees over the 30 days to October 20 came to $17.7 million, and AI agents contributed only a negligible slice of that total; just 644 agent payments had been processed on the mainnet cumulatively as of the same date. Fees flow to the validators and staking participants who secure the chain, so any future agent-driven volume would reach Ethereum (ETH) holders through the same channel as every other kind of traffic. The most realistic scenario sketched in the weekend analysis is gradual: agent payments concentrate on Layer 2 networks such as Base, where costs suit high-frequency machine transactions, and settlement on those chains lifts gas fee demand on the parent chain indirectly. Over a five-year horizon, Ethereum's capital and liquidity could pull in agents deployed on other networks, expanding transaction frequency, payment volume and the range of services sold on-chain. The same analysis concedes that meaningful fee revenue from this traffic may take years to arrive.

What survives regardless of whether Lubin's call lands is the gap between identity and commerce. More than 72,000 agents have claimed a registered presence on Ethereum, yet only 644 have ever paid on the mainnet, which means builders are staking names well ahead of revenue. The two indicators worth tracking are the ones the weekend analysis itself flags: Ethereum's share of total ERC-8004 registrations and the monthly x402 payment volume. If agents follow liquidity, those numbers rise together; if the trust layer turns out not to be the binding constraint, the registrations still cost the network little, and the $49.6 billion liquidity pool remains the deepest market an autonomous buyer could spend into.