For a long time, the first job in America came with a basic reassurance: There would be a path, it just couldn’t be a dream role. It may be repetitive, poorly compensated, or simply unrelated to what a graduate had studied. But it generally gave young people a starting point. This starting point now appears to be under pressure in the parts of the labor market most exposed to AI. A recent Stanford study, based on monthly individual-level payroll records from ADP, the largest provider of payroll software in the United States, tracked employment at tens of thousands of firms and millions of workers through September 2025. He found that the stress was falling unevenly. Among the occupations most exposed to AI, workers aged 22 to 25 saw a 6% decline in employment from late 2022 to September 2025. On the other hand, older workers in the same occupations recorded a 6% to 9% increase. Even after controlling for firm-level shocks, the researchers still find a 15 log-point drop in relative employment for young workers in the most exposed categories. The caveat here is not that jobs are disappearing everywhere at once. The market, in some corners, seems to be growing less patient with startups.
Why Fresh Graduates Are Feeling the Pinch First
New graduates are usually hired for neat, manageable bits of white-collar work – first drafts, basic analysis, routine coding, support tasks, research cleanup. The Stanford study shows that this is exactly the zone where AI is becoming useful enough for employers to start rethinking headcount. Younger workers tend to do more of this codified, repetitive work. Older workers, for all their flaws and bloated meeting calendars, are more likely to carry judgment, context, memory and a quiet practical sense that software still can’t be faked very well. So when companies go looking for “performance,” they don’t rush to cut experienced hands. They start downwards, shaving off the initial layer. This is why stress first appears in workers aged 22 to 25 years. The problem here is not the youth per se. It’s that early-career tasks—those through which people learn how work works—are becoming easier to automate, easier to redistribute, and, from an employer’s perspective, easier not to hire at all.
When AI acts as a tool, the story changes.
This is where the story becomes more important than the usual alarm about AI taking over jobs. The Stanford paper does not suggest that every profession touched by AI is turning against young workers. Instead, it points to something else: the impact depends on whether AI is being used to replace work or to support it. Where it is primarily used to automate tasks, early career employment is weakened. But in fields where it is used more to augment human activities, the pattern is less severe. The study found that the occupations with the largest estimated increases were those with the fastest employment growth for young workers. It also shows a wider distribution: about 70% of occupations in the lowest AI-exposure group recorded an increase in early-career employment between October 2022 and September 2025, compared to less than half in the highest-exposure group.For fresh graduates, this difference is not meaningful. This goes to the heart of whether or not a character still exists as an entry point. If AI helps a junior analyst work faster, a new coder test more efficiently, or handle more volume with an updated monitor, the job can survive and even improve. But when AI starts taking over easier, lower-risk work, the reason to hire an entry-level employee begins to fade. In such a case the question for an employer is no longer how to train a beginner, but whether a beginner is needed at all.This has more troubling implications here. Entry-level characters have never been valued just for quick production. They also serve as training grounds, places where competence is built slowly and sometimes inefficiently. AI, with its ability to effectively substitute, closes the deal. This tempts firms to view junior hires less as an investment in future talent and more as a cost that can be deducted.
Among AI-exposed jobs, a weak economy hits startups first.
A weak economy usually hits startups first. This is hardly a revelation. When companies get desperate, they don’t always rush to lay off. They often start more quietly than that: a small trainee batch, a pause on junior recruitment, a decision to stretch the existing team a bit. At first, this all seems like a logical business decision. The work is still being done but what is missing is the opening someone had to walk.That’s what makes Stanford so unstoppable. Researchers are asking a simple question: Are young workers doing worse simply because the market has cooled, or is something more specific going on in jobs exposed to AI? So they compare workers within the same firm over time, rather than blaming every decline on some vague economic mood. Even then, workers aged 22 to 25 in the most AI-exposed occupations still show a 15 log point drop in relative employment compared to those in the least exposed. For older workers, the pattern is much weaker.This is the real pain here. It’s not just a weak economy that weak economies do. It’s a weak economy that is becoming more selective about who it shuts down first. And in AI-exposed work, it appears to be rudimentary. Technology doesn’t eliminate entire professions, it just makes employers realize they can manage with one less newcomer. Once this instinct is established, slowness begins to decide who gets the first chance.
Keeps the paycheck, shrinks the opening.
For fresh graduates, the first warning sign isn’t a weak salary offer. The Stanford paper, in fact, notes little difference in annual base compensation trends across age groups and AI exposure levels. The trouble is showing up first: in work that isn’t quietly posted, in a junior role that sits vacant for a while or in early work that gets absorbed, spread, or handed over to software before it turns into an actual opening. Those already in may not notice much at first. However, for fresh graduates, the shift is immediate.
A job that starts out with care can end up as a lack of talent.
The US market was already less forgiving of recent graduates. Data from the Federal Reserve Bank of New York shows that unemployment among recent college graduates rose from 4.0% in Q4 2022 to 5.7% in Q4 2025. What the Stanford study adds is a sharp point: Among the occupations most exposed to AI, the ax is falling more heavily on younger workers. This suggests that the market is selectively cooling. In this scenario, the actual damage is hardly visible. This starts to show when the labor market starts eating its own future. As initial roles shrink, the damage is not limited to an unlucky graduating batch. This affects the system’s ability to produce skilled workers for the future. A first job is more than a source of income. This is where graduates become professionals through routine, supervision, refinement and time. If that layer of work is quickly automated, redistributed or stopped, firms can save money in the short term. But they also narrow the pipeline from which the next generation of experienced workers emerges. The result is a market that becomes more exclusive at the bottom and more concerned with top talent. Here’s the paradox: Employers bemoan the skills shortage even as they help eliminate the very place where skills are created in the first place.
