The metaphor is deliberately provocative: AI agents need food, and that food is tokens. Jordi Visser, the former CIO of Weiss Multi-Strategy Advisers who has spent the past few years pivoting toward crypto macro commentary, is placing a fresh bet on Ethereum—not because of DeFi yields or NFT speculation, but because he believes the coming wave of autonomous AI systems will require tokenized assets to function at scale.
The argument deserves unpacking, because beneath the surface-level buzzword collision lies a legitimate architectural question.
The infrastructure gap
Autonomous AI agents—software systems that can execute multi-step tasks without human intervention—are no longer theoretical. OpenAI, Anthropic, and a dozen well-funded startups are racing to deploy agents that can book travel, manage portfolios, negotiate contracts, and interact with other agents on behalf of users. The problem is that these agents need to transact. They need to pay for services, receive payments, and hold value in transit.
Traditional financial rails were not designed for this. A GPT-powered assistant cannot open a Chase checking account. It cannot sign a wire transfer. The compliance apparatus of legacy finance assumes a human at the end of every transaction chain. Tokenized assets on programmable blockchains, by contrast, are natively machine-readable and machine-executable. An agent with a wallet can move stablecoins, interact with smart contracts, and settle payments in seconds without human cosignatories.
Visser's thesis is that Ethereum, as the largest smart-contract platform with the deepest liquidity and the most developed tooling, will capture a disproportionate share of this demand.
The skeptic's rebuttal
The obvious counterargument is that none of this requires Ethereum specifically. Solana is faster and cheaper for high-frequency microtransactions. Centralized stablecoin issuers like Circle could build bespoke agent-payment APIs that bypass public blockchains entirely. And the timeline for meaningful agent-driven economic activity remains uncertain—current agents are impressive demos, not autonomous economic actors managing serious capital.
There is also the question of whether tokenization demand translates into ETH price appreciation. Agents transacting in USDC on Ethereum pay gas fees in ETH, but if Layer 2 solutions continue to compress those fees toward zero, the value accrual to the base layer becomes ambiguous.
Our take
Visser is right that the intersection of AI agents and programmable money is underexplored, and he is probably early rather than wrong. The more interesting question is not whether Ethereum wins, but whether the crypto industry can resist its usual impulse to financialize the thesis into oblivion before the underlying use case matures. Tokens as agent infrastructure is a serious idea. Tokens as speculative vehicles for agent-themed memecoins is the more likely near-term outcome.
