Former Google CEO Eric Schmidt got it at the University of Arizona. A real estate executive got it at the University of Central Florida, where one graduate even shouted “AI sucks!” into a microphone. A Nashville music executive at Middle Tennessee State told students to “deal with it.” The videos went viral within hours.

The anger makes intuitive sense. Eighty-one percent of Gen Z believes AI will reduce job opportunities, according to a recent Quinnipiac poll — more pessimistic than any other generation. A Gallup survey of 1,500 young Americans found enthusiasm for AI has dropped from 36% to 27% over the past year, while negative feelings have surged. These are graduates stepping into the worst entry-level job market in years, and they’ve absorbed the conventional wisdom: the robots took their jobs.

A sweeping new academic study says they’ve been pointing at the wrong culprit. The experts aren’t so sure they’ve found the right one either.

The biggest labor market study you haven’t heard of

In a working paper published this month, economists Peter John Lambert of the London School of Economics and Yannick Schindler of Oxford’s Ellison Institute of Technology analyzed 243 million new hire records and 407 million job postings across the U.S., U.K., Canada, and Australia between 2017 and 2025. The junior share of new hires — the fraction of jobs going to early-career workers — has fallen roughly 8 to 11 percentage points below 2019 baselines in every one of those countries. Entry-level hires are down 14% to 29% depending on the country.

The prevailing academic theory, echoed in viral LinkedIn posts and anxious graduation speeches, blamed generative AI. That makes sense on the surface, as this entry-level hiring collapse has coincided almost exactly with ChatGPT’s arrival in late 2022.

But Lambert and Schindler noticed something the AI-blame literature had missed. “The exposure to GenAI and WFH fall on largely the same occupations,” they write — white-collar, knowledge-intensive roles with heavy computer use. Software developers, accountants, and management consultants sit at the top of both rankings; electricians, janitors, and construction laborers sit at the bottom. “This makes plain the omitted variable problem that may surface when either shock is analyzed in isolation of the other,” the authors added in academic jargon that translated to, essentially: no one had tried to separate these factors.

When the researchers ran the exposures jointly, the GenAI effect collapsed. The remote work effect held firm. “The WFH coefficient remains negative and statistically significant in every leave-one-out replication across all four designs,” they write, “while the GenAI coefficient attenuates sharply and is often statistically indistinguishable from zero.”

Why empty offices hurt new grads

The mechanism isn’t complicated, even if the econometrics are. Firms hire junior workers not just for what they can do today, but as investments — future senior employees who will have been shaped by years of mentorship, observation, and feedback. That investment thesis depends on proximity. When a 22-year-old financial analyst can’t watch how a senior colleague handles a difficult client call, can’t absorb the unspoken norms of a trading floor, can’t get pulled into a hallway conversation that becomes a career-defining project — the value of hiring them drops. Supervisors, meanwhile, are burning time on Zoom management of people they can barely read. The math starts to favor hiring an experienced worker who needs less hand-holding.

Peter Cappelli, the Wharton management professor who has spent four years documenting the organizational wreckage of the remote-work era, arrived at the same conclusion from a different direction. Speaking to Fortune previously about his 2025 book In Praise of the Office, Cappelli and his co-author Ranya Nehmeh argued that remote work has turned young workers “very transactional” — they show up, they complete their tasks and they disengage from anything beyond a narrowly defined role. Nehmeh noted reports that companies are even offering Gen Z employees etiquette classes on how to dress for work, speak in meetings and interact with clients. These were things people used to absorb simply by being present in an office.

courtesy of Wharton School

Cappelli said that even within the Wharton school, he can see the corrosive effect at work. “I don’t need to be in the office … But I can also see how much worse the place is, because people like me are not in the office, and because we’re not in, the junior people aren’t there either, and so nobody’s there, right?” He summed it up by exclaiming, “Fine for me! … but bad for everybody else.”

The Lambert-Schindler study would seem to provide the quantitative backbone for what Cappelli has been observing qualitatively. Firms don’t just develop junior workers less under remote work — they hire them less. A two-standard-deviation gap in remote work exposure between occupations translates to roughly one in every 20 new hires shifting from junior to senior, and one in every 33 job postings raising its experience bar above three years. Cappelli, though, told Fortune that he’s not sold on this paper, as it offers “only very indirect measures of remote work and even less measures of AI use.” The study establishes a statistical correlation, he allows — “those doing remote work are also doing AI” — but it’s not obvious to him whether there is any causal relationship.

He goes further. What he hears from employers, he told Fortune, is that “AI is mainly an excuse for layoffs and not hiring, which they want to do anyway to squeeze more out of workers.” Companies paused entry-level hiring in 2022 and 2023 not because AI had arrived, he argued, but because they wanted to wait and see what AI could do before committing — and in the meantime, why hire junior workers when you’re not sure how much of their work you’ll still need? Some firms, he said, “appear to have moved off it and are just smartly giving new hires more challenging work to do.”

And yet, on the question of what actually happens inside organizations when the office empties out, Cappelli cited forthcoming research he conducted, based on 38 group interviews with roughly 760 employees across a multinational financial services company, which paints a picture that rhymes closely with what Lambert and Schindler are measuring from the outside. Managers and new hires alike told Cappelli’s team that junior employees “were lost much longer and everyone was learning more slowly.” One manager found that new hires had “a noticeable knowledge gap” three months in, compared to those been hired before remote work and he had to create a “special training program” in response A junior employee put it simply: “I have no one to casually talk with when everyone logs off. We need low-stakes interactions with senior colleagues.”

Asked what he would tell the graduating class, Cappelli offered advice that was blunt, a little cold, and probably more honest than the average commencement speech: “We are still figuring out what AI will do and how it will work. Stop listening to the pundits — their interest is in getting attention, and they do that by making extreme predictions. These technologies take a very long time to play out. There is nothing much you can do until employers start figuring out how to use them, so go worry about other things about which you can do something.”

The real villain?

But optimism requires confronting who, exactly, is benefiting from the status quo — and who isn’t.

JLL’s 2025 global survey of more than 12,000 workers put a name to the person standing between Gen Z and the career ladder: the “empowered non-complier.” Typically in their early 30s. Frequently a manager. Often in tech. The kind of worker who has enough institutional cachet that a return-to-office mandate lands in their inbox and stays there, unanswered, consequence-free. Their non-compliance, JLL found, is “less a rejection than a calculated decision based on their sense of empowerment.”

The empowered non-complier is, in other words, usually a millennial — the generation that preceded Gen Z into the workforce, accumulated the credentials, built the professional networks, and then, during the pandemic, quietly moved an hour outside the city. They bought the house with the backyard. They optimized their lives around never commuting again. And they have enough leverage that nobody is making them.

They are also, in most white-collar offices, the people that a 22-year-old new hire most needs to be near. The ones who know how deals actually get done. The ones who could answer the hallway question, pull a junior analyst into a client presentation, show by example what professional judgment looks like in practice. That person moved to Montclair or Hudson or Bucks County and is on a Zoom call right now, background blurred, wondering why their new hire seems so lost.

Cappelli added that he’s essentially booing along with the Gen Zers. “They are not in the mood to hear AI cheerleaders talking about how wonderful AI will be,” he said. “I confess I am in that group as well.  The cheerleaders and the doomsdayers are just guessing about the future, the guesses have so far been almost completely wrong, so why would you want to hear them again?”

He added that it’s a giant myth that you need “AI skills” to use LLMs, which require essentially no skills. But it does require a skill to use an LLM to achieve a certain professional outcome. “If you want to use AI to do statistical analyses, for example, yes you need some statistical knowledge but much less so that if you didn’t have an LLM.” The conclusion for Gen Z is that they have been handed a giant disadvantage, with no mentoring in the remote-work era to plug the gap. “This cohort just happens to be in the wrong place in life at the wrong time.”