You can reduce costs while reducing productivity.
Since 1980, American firms have quietly redirected the promise of automation away from progress and toward suppression — not to build faster, leaner enterprises, but to dismantle the modest economic gains of workers who had managed to earn a little more than their peers. MIT economists Daron Acemoglu and Pascual Restrepo have traced this pattern across decades of labor data, finding that the technology age delivered far less productivity than it could have, because companies too often chose to cut wages rather than expand capacity. The result is a portrait of innovation turned inward against itself, accounting for roughly half of America's growing income inequality — a reminder that tools are never neutral, only as wise as the hands that wield them.
- Firms have systematically targeted workers earning above-market wages for automation-driven elimination, treating technology as a cost-cutting weapon rather than a productivity engine.
- Workers in the 70th to 95th earnings percentile — non-college-educated people who had climbed just above the median through skill or seniority — bore the sharpest losses, their hard-won advantages erased by deliberate design.
- This strategy has been economically self-defeating: by prioritizing wage suppression over genuine efficiency, firms have offset between 60 and 90 percent of the productivity gains that automation could have delivered.
- Automation accounts for roughly half of all U.S. income inequality growth from 1980 to 2016, with one-fifth of that tied directly to the targeting of premium-wage workers — a scale of consequence most economists had not fully reckoned.
- Acemoglu argues the path forward is not to abandon automation but to recalibrate it — shifting managerial, technological, and policy priorities toward efficiency that lifts rather than extraction that narrows.
For nearly half a century, American companies have used automation not to make their operations more efficient, but to eliminate workers who had managed to earn more than their peers. A new MIT study led by economist Daron Acemoglu and Yale's Pascual Restrepo reveals this pattern with unusual precision: since 1980, firms have systematically deployed technology to remove jobs held by workers commanding a "wage premium" — extra pay earned above what comparable colleagues receive.
The finding challenges a foundational assumption about why companies automate. Rather than chasing productivity, firms have often chosen automation because it lets them cut labor costs by targeting the workers who negotiated or earned higher salaries. The consequences are vast: automation accounts for roughly half of all U.S. income inequality growth between 1980 and 2016, with about one-fifth of that stemming directly from firms replacing premium-wage workers. Those hit hardest were earners in the 70th to 95th percentile within their demographic groups — not the wealthy, but those who had climbed just above the median.
What makes this economically perverse is that it has not made American companies more productive. The researchers estimate the strategy has offset 60 to 90 percent of the efficiency gains technology should have delivered. A firm might deploy call-center automation that worsens operations overall, but if it cuts the wage bill enough, profits still rise. Acemoglu is direct: greater profitability is not the same as greater productivity. Echoing the late MIT economist Robert Solow, he notes that the computer age is visible everywhere — yet productivity statistics remain, in his word, "fairly pitiful."
The study draws on granular Census and industry data spanning 500 demographic groups across 49 industries, making visible not just how many jobs automation eliminated, but which workers were chosen and why. The pattern is unmistakable: when automation could remove a higher-paid position, firms deployed it; when the same technology might have replaced a lower-paid one, they were far less inclined.
Acemoglu, a 2024 Nobel laureate in economics, does not argue that automation is inherently harmful. Certain applications genuinely boost productivity and create cycles of growth and hiring. The problem is one of choice: managers have too often bundled productive automation with wage-suppression automation, and the latter has overwhelmed the former. If firms prioritized actual efficiency over labor cost reduction, American productivity could be substantially higher. But that would require a different understanding — and a different conscience — about what automation is truly for.
For nearly half a century, American companies have wielded automation not as a tool for making their operations leaner and faster, but as a weapon against their own workers—specifically those earning more than their peers. A new study from MIT reveals this pattern with unusual precision, showing that since 1980, firms have systematically deployed technology to eliminate jobs held by workers commanding what economists call a "wage premium," the extra pay some employees earn above what comparable colleagues make.
The finding upends a common assumption about why factories retool and offices adopt new software. We tend to imagine that companies automate to boost productivity, to do more with less, to compete harder and grow faster. The MIT research, led by economist Daron Acemoglu and Yale's Pascual Restrepo, suggests something more troubling: firms have often chosen automation not because it makes them more efficient overall, but because it lets them cut labor costs by eliminating the workers who had managed to negotiate or earn higher salaries. It is a distinction with enormous consequences.
The numbers are stark. Automation accounts for roughly half of all income inequality growth in the United States between 1980 and 2016. Of that, about one-fifth stems directly from firms targeting and replacing workers who had secured wage premiums—workers who, despite lacking college degrees, had managed to earn better pay than others in their field. The workers hit hardest were those in the 70th to 95th percentile of earnings within their demographic groups. These were not the highest earners, but they were the ones who had climbed above the median, and they became the primary targets.
What makes this pattern economically perverse is that it has not actually made American companies more productive. The researchers estimate that this misguided automation strategy has offset between 60 and 90 percent of the productivity gains that technology should have delivered. Firms have prioritized short-term profit—reducing their wage bill—over long-term efficiency. A company might adopt call-center automation that actually makes operations worse, but if it slashes labor costs enough, the bottom line improves. Greater profitability is not the same as greater productivity, Acemoglu emphasizes. You can reduce costs while reducing productivity. You can see the computer age everywhere, he notes, echoing a famous observation from late MIT economist Robert Solow, but the productivity statistics remain "fairly pitiful."
The research draws on granular data from the Census Bureau, the American Community Survey, and detailed industry statistics, analyzing 500 demographic groups across 49 industries. This allowed Acemoglu and Restrepo to track not just how many jobs automation eliminated, but which workers bore the brunt and why. The pattern is unmistakable: when a firm could automate away a job held by someone earning above-market wages, it did. When the same technology could have eliminated a lower-paid position, firms were less inclined to deploy it.
The implications ripple outward. Automation has widened the gap between rich and poor more than most economists realized. It has also meant that American workers without college degrees who had managed to secure decent-paying jobs—perhaps through union membership, seniority, or skill—found themselves expendable. The technology was not neutral. It was aimed. And it worked.
Acemoglu, who shared the 2024 Nobel Prize in Economic Sciences, is careful not to argue that automation itself is bad. Certain types of automation can genuinely boost productivity and create a virtuous cycle in which firms earn more and hire more workers. The problem is that managers face a choice, and they have often chosen wrong. They have bundled genuinely productive automation with the wage-suppression kind, and the latter has overwhelmed the former. The study suggests that if firms calibrated their automation more carefully, prioritizing actual efficiency gains over labor cost reduction, American productivity could be substantially higher than it is today. But that would require a different set of priorities—and perhaps a different understanding of what automation is actually for.
Citações Notáveis
There has been an inefficient targeting of automation. The higher the wage of the worker in a particular industry or occupation or task, the more attractive automation becomes to firms.— Daron Acemoglu, MIT economist
We could be missing out on potentially even better productivity gains by calibrating the type and extent of automation more carefully, and in a more productivity-enhancing way. It's all a choice, 100 percent.— Daron Acemoglu
A Conversa do Hearth Outra perspectiva sobre a história
So the study is saying companies didn't automate because the technology was better. They automated to get rid of expensive workers.
More precisely, they automated in ways that targeted workers earning above-market wages. A firm might have had five options for where to deploy a new technology, but it chose the one that eliminated the highest-paid workers, even if that wasn't the most efficient choice overall.
But wouldn't a company want to be as productive as possible? Isn't that how you stay competitive?
You'd think so. But in the short term, cutting your wage bill improves your quarterly numbers. A manager can show higher profits even if the company is actually less productive. Those are two different things, and the incentives often point toward the latter.
What happened to the workers who lost those jobs?
They either left the workforce, took lower-paying jobs, or stayed in the same field but at reduced wages. The study focused on non-college-educated workers who had managed to earn decent pay—union workers, skilled tradespeople, people in the 70th to 95th percentile of their field. They were the ones most targeted and most harmed.
And this affected inequality how?
Automation accounts for about half of all income inequality growth since 1980. One-fifth of that comes specifically from this wage-suppression strategy. So firms didn't just fail to lift workers up with new technology—they actively used it to push certain workers down.
Is the argument that we should have less automation?
No. The argument is that we're using automation badly. We're optimizing for short-term cost reduction instead of long-term productivity and growth. If firms made different choices about where and how to automate, we could have both better productivity and less inequality.