Cathie Wood tells investors to “follow the agents” as AI starts spending money
Her Robinhood Summit framing spotlights a looming fight over whether agent payments run on open stablecoin rails or closed stacks.
ARK Invest CEO Cathie Wood said investors may need to start “following the agents” as AI systems shift from answering questions to executing purchases and payments. The investable question, she argued, is which rails those agents will use when they start spending real money at scale.
Cathie Wood’s “Follow the Agents” Heuristic Lands on Wall Street’s AI Trade
Cathie Wood used a familiar investor shortcut in a new place this week, and the nuance matters. Speaking on a panel at Robinhood’s Summit in Houston on Wednesday, the ARK Invest CEO said, “We’re probably going to be talking more and more about ‘follow the agents,’” framing AI agents not just as a product category but as a new source of observable demand.
Wood has long told investors to “follow the developers” to see where technology is headed, on the logic that engineers cluster around tools that work. Her update shifts the lens from builders to behavior: if AI agents increasingly choose which software, services, and networks to use, then their spend becomes a measurable signal that can show up in transaction flows rather than in narratives.
The catch is that the packet offers no adoption metrics yet. There are no disclosed volumes, transaction counts, or market-share splits for agent-driven payments, which keeps Wood’s heuristic in the “framework” bucket for now. Still, it cleanly reframes the AI trade into an infrastructure contest that traders can track once the data starts to leak out.
Open vs Closed Rails: Who Gets to Process Agentic Commerce?
The open question underneath Wood’s line is straightforward and market-relevant: as AI agents begin spending real money, will those payments settle on open networks like stablecoins and public blockchains, or will they be routed through closed systems controlled by a small number of banks, payment providers, and large technology platforms.
Joseph Chalom, co-CEO of SharpLink and former head of digital assets at BlackRock, argued last month that the agentic financial system should not consolidate inside a handful of corporate stacks. “A world full of intelligent agents means nothing if a handful of companies decide where your money can go,” he wrote in the final installment of a three-part series on agentic finance.
Chalom’s argument is less about token price reflexes and more about the mechanics that make agent spending tolerable to users and regulators: permissions, limits, revocation, audit trails, and portability. He gave a concrete example of a user authorizing an agent to spend up to $500 to book a hotel without granting unlimited access to a bank account, and said users should be able to cancel that authority and see a record of what the agent did.
Portability is the other wedge. Chalom argued people should be able to move their agents between financial providers rather than getting locked into one company’s system, with the agent carrying identity, financial information, and permissions in the way a phone number can be ported between carriers. In his framing, open blockchains such as Ethereum could provide a common financial network for agents, apps, and companies, letting money move across shared rails without a single bank or technology company intermediating every transaction.
The Early Breadcrumbs: Ethereum as Shared Network, Stablecoins for 24/7, and Coinbase’s Agentic-Trading Hint
BlackRock made the machine-payments case more explicitly in a September paper on the overlap between AI and digital assets. The asset manager argued AI agents may need to pay for API calls, buy data, or rent computing power without waiting for a person to approve each transaction, and positioned stablecoins and blockchains as a fit for that pattern because stablecoins can move 24/7 and blockchain-based payment protocols can support small software-to-software transfers.
The paper also referenced Coinbase’s x402 as a design intended to let machines pay for online services such as data or API access, which is the kind of plumbing detail that would turn “agent commerce” from a concept into an instrumented flow traders can actually measure.
There is also a thin but notable anecdote from inside crypto market structure. Coinbase CEO Brian Armstrong wrote on X that “Grok is the leading client for agentic traders on Coinbase currently,” without providing figures or further details, leaving the scale and composition of that activity unresolved.
Competition for agent checkout is not limited to crypto-native rails. Stripe, Visa, Google, and OpenAI were named as companies developing ways for agents to make purchases, and BlackRock’s framing left room for traditional payment systems to remain important, which sets up a realistic mixed-rail outcome rather than a clean winner-take-all story.
The next tranche of signal is going to be data, not more slogans: any disclosed metrics that separate agent-driven payments on stablecoins or public chains from card and bank rails, follow-on detail from Coinbase on what “agentic traders” means in volume or user terms after Armstrong’s Grok comment, and product announcements from Stripe, Visa, Google, or OpenAI that reveal whether agent commerce is being routed through closed platforms or interoperable networks.
My Take: The Trade Isn’t ‘AI’—It’s Where the Spend Settles
The filing-cabinet version of this story is that Wood offered a new heuristic, but the market version is that she pointed at a future dataset. “Follow the agents” only becomes tradable when it translates into observable settlement behavior, and right now the packet is explicit about what’s missing: no volumes, no transaction counts, no share-of-rail breakdown.
The threshold that matters is whether agent payments for machine-native goods like API calls, data, and compute start to show up as repeatable stablecoin and onchain settlement flow rather than staying abstracted inside closed bank and Big Tech stacks, because that is what turns agentic commerce from a theme into a measurable demand driver for specific rails.
