How a Yale Economics Professor Predicted Today’s Job Market Collapse Back in 2023

Pascual Restrepo joined the Yale economics department in the spring of 2023, when ChatGPT was still being discussed as a novelty and most economists were cautioning against overreaction. He started mapping the specific mechanisms by which AI would hollow out the labor market, not in a general way but in precise, task-level detail that made his predictions harder to dismiss as speculation. Since January 2023, the number of entry-level job postings in the United States has decreased by 35%, which is precisely the cohort of roles that his research identified as most exposed. As a result, that work is currently being cited in corporate boardrooms, financial media, and policy circles.

Having spent years working with Daron Acemoglu at MIT, the economist who won the 2024 Nobel Prize in Economics in part for his research on how technology changes labor markets and wages, Restrepo approached this with a specific perspective. Their common framework is based on a conceptually obvious but often misinterpreted observation: workers are not replaced by automation as most people think. Jobs are not completely eliminated by it.

Because they are typically the predictable, codifiable tasks that produce the most economic value and are therefore the most worth automating, it eliminates certain tasks within jobs, frequently the ones that pay the highest. The premium pay associated with those tasks vanishes along with them. The employee is repriced rather than fired. And the kind of wage compression and growing inequality that Restrepo and Acemoglu documented—arriving long before most people were aware of it—is driven by this repricing, which is taking place across millions of jobs concurrently.

The most lucid version of the story is presented by the entry-level data. According to analysis by labor research firm Lightcast, postings for entry-level positions in the US decreased by 35% between January 2023 and June 2025. This decline occurred prior to the widespread implementation of AI in most workplaces and is more indicative of a structural issue than a slowdown in hiring.

Months before ChatGPT’s launch, in early 2022, the Budget Lab at Yale discovered that conditions for jobs exposed to AI had already started to deteriorate. In November 2025, CNBC reported that AI was specifically targeting young workers’ career advancement pipeline by removing the kind of early professional experience that people in their twenties have traditionally used to develop expertise. The data is starting to support Restrepo’s framework, which predicts a quiet, gradual erosion of the economic on-ramps that earlier generations used to gradually accumulate skills and wages rather than a dramatic, abrupt disruption.

The analysis is pushed into areas that most economists still approach cautiously by the more recent and provocative extension of Restrepo’s work. In a paper titled We Won’t Be Missed: Work and Growth in the AGI World, he outlines the consequences for an economy in which AI is able to carry out nearly all economically significant tasks. He contends that in that case, people are no longer the economic production bottleneck. The limited resource is processing power. Whoever controls the hardware to run AI systems would drive growth, and labor’s share of total economic income may eventually converge toward zero—not because workers are fired, but rather because the jobs that generate income no longer require them. Restrepo links this scenario to the 2030s, which is closer than most people would think.

Restrepo’s emphasis on a subtlety that is often overlooked in the headlines sets him apart from the more alarmist voices in this discussion. He told a Yale audience recently that “people have the wrong intuition when they say that if AI can do my job for ten dollars an hour, then my wage falls to ten dollars and my life is terrible.” According to him, the crucial factor is what ten dollars can purchase.

Workers with lower nominal dollar incomes may still be able to maintain or increase their purchasing power if AI lowers the cost of goods and services in a wide enough range of industries, such as research, healthcare, education, and transportation. A straight-line wage collapse is not the same as a world where AI competes your salary down while also lowering the cost of rent, groceries, and childcare. In his more pessimistic scenario, AI only impacts specific job categories while prices in other sectors of the economy stay the same. In this scenario, displaced workers bear the full brunt of the pressure on their wages to decline without any compensation.

Which of those outcomes will happen is still unknown. Together, the research from MIT, the Budget Lab at Yale, and Restrepo’s own work indicates that the shift is already taking place—not waiting for AGI or the next generation of language models, but happening right now through task-level automation that is gradually repricing the most codifiable aspects of knowledge work. The youngest workers are the ones who are most affected because they are entering a job market that offers them more uncertainty and fewer starting points than any other cohort since the 2008 financial crisis. The economist who gave this mechanism its most accurate name in 2023 is seeing it manifest in the data, and the predictions he made about it are starting to uncomfortably appear to be accurate.