Money Lags Knowledge in AI

The gap

Four or five years into the AI era, AI has effectively eaten knowledge work and has barely touched money. Micky Malka raises this as the acknowledged gap in his own investment thesis: "Today, token factories don't touch money that much. Money has been outside of AI, and we're on year four or five. It's funny because money is supposed to be one of the first most fungible things that should be playing the game, and it's not."¹ Elsewhere, more plainly: "we're behind compared to knowledge. Knowledge went much faster. Money's behind, which is fine."

Why money should have moved first

Malka's method is to follow flows, which is why he expected money to be an early rather than a late mover: "Nothing is static. Money is never static. Money is like gravity. It always falls. It always moves. It's always going somewhere. It's like water. It's never stuck." His argument is that fungibility combined with constant motion should have made money the easiest substance for machines to work with, and instead it has been the last.

What closes the gap

His answer is the intersection of blockchains and AI specifically: you will need to have intelligent smart contracts that will do things on your behalf. Not AI merely reading financial data, and not blockchains merely moving value, but agents executing directly on programmable rails. The builders he names working on this are established, founder-led incumbents rather than startups, including agentic payments work at Stripe, where Malka notes John Collison has described it as effectively the only thing the company is working on, and tokenized equities work at Robinhood.

The first visible signal, in his account, is not a payments company at all. It is prediction markets, which some call gambling and others call a place to bet on anything around the clock, and what matters to him is specifically the around-the-clock part, because agents need rails that never close. Markets that never close are, in his framing, the first consumer-facing infrastructure built to a machine's schedule rather than a bank's.

Why it matters

If the claim holds, it locates the next few years of value creation precisely: not in more capable models, but in wiring existing capability to the value layer. It is a bet on plumbing over intelligence. It also reframes what AI applied to fintech has meant so far, which has mostly been AI applied to money as a subject, underwriting, fraud detection, support, analysis, all knowledge work about money. Malka is describing something categorically different: money as a token type that agents transact in directly, the difference between a model that reads a statement and an agent that actually moves a balance.

Tensions and open questions

The claim is asserted rather than sized: there is no accounting of how much of the money stack is already AI-touched, and no definition of what would count. That runs against a broader body of evidence documenting substantial building already underway on the money side of AI, so a more defensible reading may be that consumer-facing financial decisioning specifically lags, not that infrastructure does, though that reconciliation is not one Malka himself offers. The framing that money should be easy because it is fungible also inverts the more obvious explanation: money moves slowly into AI precisely because it is regulated, irreversible, and adversarial, not because anyone forgot to build the rails, and Malka, who understands this as well as anyone, does not address it directly. Why the specific fix is smart contracts on blockchains, rather than existing rails paired with better authorization, is also asserted rather than argued.