In late February of this year, the American payments company Block announced it would shrink its workforce from 10,000 employees to fewer than 6,000 — a cut of nearly half. Founder Jack Dorsey said the reason was not financial trouble but that artificial intelligence (AI) tools now let a smaller team do far more work. The market applauded. Block’s shares jumped around 20 percent in the hours after the announcement.
For a single company, the move is rational. Labor costs fall, profits rise, the stock climbs. The trouble starts when every company runs the same arithmetic — because the worker you lay off is another company’s customer. When jobs disappear, spending shrinks; when spending shrinks, those companies’ revenues fall too.
Economists at the Wharton School call this structure the “AI layoff trap.” A firm that cuts staff pockets the full savings, but the drop in demand it sets off is shared across the entire market. So each firm races to cut headcount before its rivals do, and society as a whole automates faster than it should.
The money that used to flow to wages does not vanish into thin air, though. It shifts to shareholders and to capital. So the real question is not whether demand disappears but who receives it. Higher earners tend to spend a smaller share of what they make and to save the rest, so even with the same national income, when it pools at the top, overall consumption sags.
It is also far from settled that the technology is as capable as advertised. An economist who won the 2024 Nobel Prize in Economics estimates that AI’s effect on productivity will come in under one percent over the entire next decade. And last year most companies reported no clear change in employment or productivity despite enormous spending on it. As one Wall Street economist put it, AI is everywhere except in the economic data.
Earlier waves of automation usually targeted a single task. The power loom replaced hand weaving; the spreadsheet replaced manual calculation; the rest of the work stayed with people. What sets general-purpose AI apart is that it takes aim at office and professional workers across many industries at once. Accounting, analysis, junior legal work, radiology, software development — roles long thought safe from automation — are now on the table together.
History also shows that when new technology displaces labor, the adjustment takes a generation or more. In Britain during the Industrial Revolution, wages and employment took decades to climb back to where they had stood before the new machinery arrived. In the United States, farm labor fell from 90 percent of the workforce to 2 percent over the course of 140 years. That entirely new kinds of jobs emerged in the meantime is some comfort, but as one economist observed, the short run can last a lifetime.
The unease is already surfacing as friction in the real world. This year in the United States, someone fired shots at the home of a city councilman who had approved a data center, and a firebomb was thrown at the home of an AI company’s chief executive. It is the picture that forms in the gap between the future the technology promises and the present that people actually feel.
In the end, what matters most in this shift is less how smart AI becomes than how a society divides the gains it produces — and that is where the center of gravity of the current debate now sits.