College graduates’ dread about the job market is so acute that commencement speakers who dare mention artificial intelligence keep getting booed. The jeers are not hard to understand. Those building AI have been candid, if not always consistent, about their ambition to automate large categories of white-collar work, including much of what a college degree has historically unlocked. A year ago, Anthropic CEO Dario Amodei warned that “half of all entry-level white-collar jobs” could disappear within one to five years and that unemployment could spike to “10 or 20 percent.” Lately, he and OpenAI CEO Sam Altman have sounded less apocalyptic, conceding that the wipeout may not arrive after all. The whiplash is hardly reassuring. Anyone who has recently applied for a job has felt the process becoming more automated and less human. The grim mood is supported by data: Entry-level job postings are down in some fields, and unemployment among recent college graduates is creeping up. The share of the long-term unemployed with bachelor’s degrees was one in five a decade ago. Now, it’s closer to one in three.
The case for doom flatters nearly everyone who repeats it. For journalists on deadline, it has the value of being easy to illustrate. There is always a recent graduate, somewhere, willing to be photographed with a sullen face in a sunlit childhood bedroom beside a laptop full of rejected applications. AI companies need the world to believe their progeny is powerful enough to wipe out categories of work. The AI panic helps sustain hype-driven stock valuations that have defied gravity. And when apocalypse talk becomes inconvenient—for example, as OpenAI and Anthropic reportedly prepare for blockbuster IPOs—the same executives can soften their Cassandra tone into a homily about history’s long record of technological adaptation, just as they are doing now.
The way to understand the AI moment is not that it makes college worthless or that white-collar work will collapse. The entry-level slowdown predates ChatGPT, but AI appears to be accelerating what Laura Ullrich, the lead economist at Indeed, the job service, calls “experience creep”: employers demanding more experience for jobs once available to candidates still acquiring it. The labor market still wants college graduates. It increasingly wants them with the experience and critical-thinking skills that entry-level jobs used to develop.
Since ChatGPT arrived in 2022, America’s economy has added about 3 million white-collar jobs, while blue-collar employment has remained flat. The occupations often cited as AI’s roadkill have instead expanded: there are 7 percent more software developers than in 2022, 10 percent more radiologists, and 21 percent more paralegals. Real wages in professional and business services are up 5 percent; among office and administrative workers, 9 percent. Controlling for the usual variables, white-collar workers now earn roughly a third more than blue-collar ones—almost triple the premium of the early 1980s.
It may be too early for the damage to show up in the data. Perhaps. But there are reasons to doubt a looming doomsday. Companies need people to train the models, adapt them to specific businesses, determine when they are wrong, and prevent them from doing stupid or dangerous things at scale. The era of autonomous agents is already producing a small army of humans to attend to AI and make sure it behaves.
White-collar work has rarely been as automatable as those selling automation insist it is. The British advertising executive and behavioral economist Rory Sutherland calls this the “doorman fallacy”: the mistake of assuming that because a doorman opens the door, opening the door is the job. But a doorman also recognizes residents, screens strangers, receives packages, hails taxis, notices when something is awry, and maintains the quiet order of a building. A junior analyst, likewise, does more than assemble spreadsheets. Jobs are more than tasks; they are bundles of skills, habits, relationships, judgments, and institutional knowledge. New technologies tend to eat the bundle unevenly.
But if machines eat the bundle unevenly, the consequences will not be evenly distributed, and recent graduates are not wrong to worry. The labor market for white-collar work, even entry-level white-collar work, can be healthy, even as the first step becomes a harder climb. Much of the easily automatable work of the junior job is being absorbed into the AI models. This leads employers to want entry-level applicants to have more experience than in the past, and the savvy to complete more difficult work.
This dynamic is perfectly represented in a new report about entry-level hiring. The Strada Institute for the Future of Work (which has been a supporter of the Washington Monthly) asked nearly 1,500 executives and senior talent leaders what AI is doing to their junior ranks. Among employers that have begun using the technology, 46 percent said it had increased their entry-level hiring over the past year. Only 13 percent said the opposite. The junior worker, in other words, is not disappearing.
What is disappearing is the easy work of the junior worker: 42 percent of executives said AI has expanded the analytical and judgment-based work asked of entry-level employees, while 41 percent said it has stripped away routine and administrative tasks. The tasks a new graduate used to cut their teeth on—the photocopying, the first-pass research, the formatting and summarizing—are what AI does best. What remains is the harder part.
“What used to be the entry level—these kinds of transitional roles, where you do a bunch of grunt work, prove yourself, and then learn on the job to prepare for the next level—that’s going away,” Andrew Hanson, Strada’s research director and the report’s lead author, told me. “Entry-level roles are becoming more like mid-level roles.”
When the executives were asked to rank the skills they sought, their list read like a brochure for a small liberal arts college. Critical thinking came first, followed by communication, collaboration, workplace readiness, and self-management. But when employers were shown profiles of recent graduates and asked whom they would hire, the candidate with a 4.0 grade-point average, academic honors, and no work experience came in last. Coming in first were the candidates with internships or other industry experience. Employers still want the qualities colleges say they teach. They’re less willing to accept the transcript alone as proof.
Data from Indeed reflects this shift: From mid-2022 to mid-2025, the share of tech job postings open to workers with two to four years of experience fell from 46 percent to 40 percent, while the share seeking at least five years of experience rose from 37 percent to 42 percent.
“The whole idea that you need work experience for an entry-level job seems really counterintuitive,” Hanson said. He thinks part of what’s happening is that academic achievement has become a weaker signal. “With grade inflation, academics are not the differentiators they used to be,” he told me. Employers have gone looking for a new one, and “that tends to be the work experience,” especially internships and other work-based learning.
That does not mean college has stopped paying off. The data say otherwise. But AI is accelerating a shift that leaves the four-year degree valuable and insufficient at the same time. Unless all colleges start reliably giving students real work experience before they finish, many graduates will find that getting their first job now requires proof that they have already had one. For the class of 2026, that is something no commencement speaker can explain away. Until it changes, the students will keep booing.
