A March 2026 essay by economist David Oks is making waves in AI labor debates by correcting a popular political talking point. When J.D. Vance told the New York Times that America has 'more bank tellers today than when the ATM was created,' he was citing a well-worn parable. The problem, Oks notes, is that it stopped being true around 2010. Teller employment has since collapsed — not because of a better ATM, but because of the iPhone.
The distinction matters more than it might seem. ATMs automated specific tasks but left the branch model intact. Because that model was built around human roles, it generated its own demand: the Jevons effect meant banks opened more branches as costs fell, actually increasing total teller headcount even as each branch needed fewer. Mobile banking didn't optimize the branch — it made it optional. That's a different kind of disruption, and it arrived thirty years after the ATM with almost no warning.
Oks draws on work by economists David Autor, Daron Acemoglu, and James Bessen to make a broader claim: task automation inside human-shaped systems tends to preserve employment through complementarity. The legal research market offers a live example — AI tools are picking off discrete tasks, but workflows are largely intact. The more interesting question is whether some product eventually makes the underlying workflow itself redundant, the way Quicken didn't kill bookkeepers but cloud accounting software is quietly doing so.
Oks stops short of predicting imminent mass job loss. His argument is more unsettling than that: the absence of short-run disruption is not a verdict. People watched ATMs roll out for three decades and concluded teller jobs were safe. They were — until a product came along that made the whole question moot.