AI Labor Tracker / Labor

The grad job market looks normal until you sort by major

New college graduates had a normal-looking summer. Underneath the average, computer science grads took a hit the size of a recession, and a degree bought less protection than it used to.

The AI jobs crash for new graduates didn’t show up in the averages this year. It showed up in specific majors, and the people hit hardest were the ones who did exactly what we told them to do: learn to code.

That’s two claims, and both are backed by data released this fall. They aren’t in conflict. They just measure different things.

The upside is worth stating plainly. Most of the class of 2026 found the same job market its older siblings did.

And in the Census data covered below, nursing, teaching and civil engineering graduates saw entry outcomes improve. The question, as always, is who loses the work, and what we owe them.

The average looked fine

Start with the headline number. Economists Robert Fairlie and Jane Wu at UCLA used the Current Population Survey to track unemployment among college graduates ages 22 to 25 who aren’t in school. Their IZA discussion paper is an early read on the class of 2026’s first summer.

Averaged over June, July and August, unemployment for that group was 7.3% in 2026. It was 7.1% in 2022, 6.3% in 2023, 7.8% in 2024 and 7.2% in 2025. So this summer was worse than three of the last four, and better than 2024.

There’s a fair objection to the official rate. It leaves out people who have given up looking.

Fairlie and Wu built an expanded measure that counts these “sidelined” graduates. On that basis, summer 2026 hit 10.4%, the highest of the five summers, though it beat the previous high by only 0.3 percentage points.

Summer always looks bad, because fresh graduates flood the market at once. The rate for this group was 5.3% in winter 2026. It then jumped from 5.4% in May to 7.8% in June, before easing to 6.8% by August.

Compared with older graduates ages 30 to 49, the gap was unchanged at 4.5 points. Two broader gauges did rise.

A proxy for underemployment, which adds graduates working under ten hours a week, was 9.1% this summer against 8.2% in 2022. A mismatch measure, which adds graduates in jobs needing no formal credential, was 17.1% against 15.6%.

The authors also tested whether graduates headed for more AI-exposed occupations fared worse this summer. The estimates pointed that way but weren’t statistically significant.

As the UCLA Anderson Review summarized it, the jobpocalypse “didn’t arrive.” Earlier in the year, BlackRock CEO Larry Fink had warned that this class could see “the highest unemployment rate among them in years.” On the standard measure, that didn’t happen.

The degree’s edge narrowed

One comparison in the Fairlie and Wu paper deserves more attention than it got.

Young people of the same age without a college degree saw unemployment fall, from 7.5% in summer 2025 to 6.8% in summer 2026. Young graduates rose slightly, to 7.3%. That leaves graduates about half a point worse off than non-graduates on the official rate.

On the expanded measure the college edge shrank from 2.2 points to 0.3, the smallest of the five summers. The two groups were nearly even.

It’s one summer, and the authors say the comparison deserves more study. Both groups are the same age and face the same economy, which is what makes the shift interesting. It’s a strange result for a credential that’s supposed to buy safety.

The majors that did crash

Averages hide what happens to particular groups. Two studies published in September show that, and both look at classes that graduated before 2026, which matters for reading them.

The sharper one comes from the Census Bureau. Researchers Cody Orr, Lee Tucker and Lawrence Warren used the Post-Secondary Employment Outcomes data, which links college transcripts to earnings records. It covers roughly 29% of the bachelor’s degrees awarded in the US from 2016 to 2024.

They scored each major by how exposed its graduates’ typical jobs are to large language models, using a widely cited GPT-4-based measure of task exposure. Then they watched what happened after ChatGPT arrived in late 2022.

The results are blunt. For the most exposed tenth of majors, the likelihood of initial employment fell by five percentage points. Full-quarter initial earnings fell by 13%.

The authors write that this earnings loss “is comparable in magnitude to the earnings losses associated with graduating into a large recession.” That’s a recession for some graduates, while the rest had a normal market.

Here’s how it looks by major. The numbers below are the paper’s own estimates for 2024 graduates, measured against 2022 and against the average change across all majors.

Computer science graduates were 6.1 points less likely to be employed in the quarter after graduation than the average change across majors would suggest. Their earnings were about 19 log points lower, around 17% in plain terms.

Computer engineering lost 6.8 points of employment. General computer and information systems lost 5.4.

Setting aside a couple of very small fields, the authors note that these three majors had both the largest earnings declines and the largest employment declines. Because the earnings measure uses full quarters of work, the drop isn’t just people starting later in the quarter. Fewer of them found jobs, and those who did were paid less.

At the other end, nursing graduates were 2.8 points more likely to be employed, with earnings about 3.9 log points higher. Elementary education, criminal justice, civil engineering and medical assisting also gained.

The middle is mixed. Accounting barely moved, and its earnings rose. Finance lost two points of employment.

Marketing and communications lost some earnings but kept their jobs. Journalism, in the top decile, lost about a point of employment and 6.6 log points of earnings.

Where the coders went

The detail that stays with me is where these graduates ended up instead.

The Census team found that about half the earnings drop came from lower pay inside the same industries. The rest came from a shift into worse-paying ones.

Graduates in the most exposed majors became more than two points more likely to work in restaurants and hotels, and similarly more likely to work in retail. Those are two of the lowest-paying sectors for these graduates.

Meanwhile the information sector’s share of these graduates fell by over three points. Professional, scientific and technical services, their largest source of high-paying jobs, shrank by nearly six.

Some went elsewhere. Using a separate Census survey, the authors found self-employment in the top decile rose from 1.01% to 1.49%, and graduate school enrollment from 16.20% to 17.86%.

Taken at face value, those moves could explain up to half of the five-point drop. They also warn that people may be choosing those paths because the jobs aren’t there.

Time helps, but slowly. A year out, graduates in the seventh, eighth and tenth deciles had recovered nearly half their initial employment losses. The top decile was still about 4.2 points behind the least exposed majors.

Texas tells the same story

The Dallas Fed looked at Texas administrative records with a different exposure measure. Samuel Dodini and Tucker Smith built their score from Anthropic’s records of the tasks its Claude platform performs, matched to the occupations graduates enter.

Their finding: a major with 10 percentage points more automatable tasks saw a 1.7-point relative decline in the chance of finding work in Texas within a year. Graduates from more exposed majors who did find work earned about 5% less, relative to less exposed majors, from 2021 to 2024.

The most exposed majors include computer science, computer engineering and languages. Among the least exposed are nursing, education and psychology. That’s the same ranking as the Census study, from different data and a different exposure score.

The Dallas team also found students reacting. May 2024 graduates from more exposed majors were more likely to head to graduate school, and two-thirds of 2024 computer science graduates who went back chose computer science again.

But new master’s holders in exposed fields saw earnings drop about as much as bachelor’s graduates. More school in the same field didn’t buy much.

Undergraduate enrollment is already shifting. For each 10-point difference in a major’s automatable task share, enrollment at Texas four-year schools fell 4.8% between fall 2024 and fall 2025.

The risk in this reading

The obvious risk is blaming AI for something else. Tech hiring cooled after the pandemic boom, and computing majors would have felt that regardless of any chatbot.

The studies take that seriously. The Census paper compares majors rather than years, so a general slowdown would hit everyone.

Its authors also note that 2022 earnings were already depressed, which makes some middle-decile results harder to read. And the UCLA write-up flags remote work as a possible factor: unemployment was higher this summer among recent graduates linked to jobs that can be done remotely.

Even so, the pattern lines up with exposure across more than a hundred majors, in two separate datasets, with two different measures. Finance and journalism, far from Silicon Valley, lost ground too. That’s hard to explain as a tech-sector story alone.

There’s a risk on the other side too, of reading the averages and calling it a false alarm. New York’s business community is moving fast.

The Partnership for New York City reports that AI companies leased more than 2.2 million square feet of office space in the city in the first half of 2026. It also cites a poll in which more than half the class of 2026 said they need stronger AI skills.

My read: the averages are telling the truth about the market and lying about the people. A normal unemployment rate can sit on top of a recession-sized loss for one group of 22-year-olds.

Dignity of work, with or without a degree

This connects to something I’ve written about before. In the job thermometer, the postings data showed entry-level screen work cooling fastest. These studies show who sits on the other side of those postings, by major, and the reading is cold for some and warm for others.

It also connects to work that doesn’t need a degree at all. Young people without one had a better summer than graduates on the official measure. The trades that build and run the data centers behind these AI systems keep hiring, as I wrote in the data center electrician.

None of that means college was a mistake. Nursing and teaching graduates did better than before. It means the four-year degree is no longer a blanket guarantee, and the jobs that were never supposed to need one deserve the same respect and pay.

When a computer science graduate ends up in food service, the problem is a ladder that lost its bottom rungs, with no second ladder built beside it. Lazy kids have nothing to do with it.

What to do now

If you’re choosing a major, look at what the major’s graduates actually did in 2024, not what the field promised in 2019. The Census table lists more than a hundred majors. Health, education and hands-on engineering held up well.

If you’re already in a computing program, don’t count on a master’s degree in the same field to bail you out. The Texas data says it didn’t. Pair the degree with something closer to the physical world, such as health systems, infrastructure or skilled-trade work.

And if you run a university or a workforce board, publish your own outcomes by major every year, the way these researchers did. Students in Texas are already shifting majors on thin information. Give them better information, sooner.

Sources

1. “The Early Impacts of AI on Employment among Recent College Graduates,” Robert Fairlie and Jane Wu, IZA Institute of Labor Economics, IZA DP No. 18945, Sep 2026. https://docs.iza.org/dp18945.pdf
2. “For New College Grads, the AI Jobpocalypse Didn’t Arrive,” UCLA Anderson Review, Oct 7, 2026. https://anderson-review.ucla.edu/for-new-college-grads-the-ai-jobpocalypse-didnt-arrive/
3. “Graduating into Disruption: Labor Market Outcomes for AI-Exposed College Majors,” Cody Orr, Lee C. Tucker and Lawrence Warren, U.S. Census Bureau, CES 26-56, Sep 2026. https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-56.html
4. “AI plays a role in weak labor market for college graduates,” Samuel Dodini and Tucker Smith, Federal Reserve Bank of Dallas, Sep 22, 2026. https://www.dallasfed.org/research/economics/2026/0922
5. “Artificial Intelligence Is Already Reshaping New York City,” Partnership for New York City, Oct 2, 2026. https://pfnyc.org/research/artificial-intelligence-is-already-reshaping-new-york-city