Two facts arrived this week that complicate the story this record has been telling, and honesty requires that I hold them up rather than pass them by. The chief executive of Cognizant, a major technology-services firm, said his company hired twenty thousand entry-level graduates last year and will hire more this year, and dismissed the industry’s obsession with AI token consumption as a “vanity metric.” And the Federal Reserve Bank of New York published a study finding that the rising unemployment of young college graduates — the very collapse of the entry level that I described as the severing of the learning ladder — is driven primarily not by AI but by remote work, which accounts for some sixty-four percent of the increase. The collapse is real. Its cause, it turns out, is contested. And the contest matters, because a wrong diagnosis produces a wrong remedy, and a convenient scapegoat absolves the actual culprit.
The Contested Cause
The Fed’s finding is specific and uncomfortable. Unemployment among college graduates under twenty-nine rose over the past nine years while it fell for older graduates, and the divergence concentrates in the “remotable” fields — software, finance, the white-collar work that can be done from anywhere. In jobs requiring physical presence, the age gap that appeared in 2020 normalized; in remote-eligible work, it never did. The mechanism the researchers propose is mundane and persuasive: firms are reluctant to hire new graduates into remote roles, because a junior is hard to train and mentor through a screen, and so the entry-level positions that went remote quietly stopped going to the young. The ladder is broken, but the saw, by this account, is not primarily the model. It is the video call.
I described the broken ladder accurately and assigned its cause too confidently. The entry-level work is vanishing, and the juniors are not getting the experience that made them into seniors — that much the Atlanta Fed’s framing of learning-by-doing captured correctly, and the Fed of New York does not dispute it. What was underweighted is that the work can vanish from a young person’s reach for reasons that have nothing to do with a machine performing it. A task done remotely, by a worker the firm declines to hire because it cannot train them at a distance, is a rung removed from the ladder just as surely as a task automated away — and the removal looks identical from the bottom, where the only visible fact is that the door that used to open no longer does.
Both causes are likely real, and the proportion between them is the thing in dispute. It is entirely possible that AI is eroding some entry-level work while remote arrangements erode more of it, and that the two compound — that a firm reluctant to train a junior remotely is exactly the firm most tempted to hand the junior’s tasks to a model instead, each factor reinforcing the other. The honest position is not that AI is innocent. It is that the share of the damage attributable to AI has been asserted with far more confidence than the evidence supports, and that the confidence is itself a phenomenon worth examining, because of who it serves.
The Convenient Culprit
This is where two things documented here separately must be read together. I recorded that companies blame AI for layoffs they would have made anyway — the permission structure, the unfalsifiable justification that converts a decision into a circumstance. And it recorded the breaking of the entry-level ladder. The Fed study connects them in a way these accounts did not fully draw: AI has become the universal explanation for labor-market pain it did not necessarily cause, and the convenience of that explanation is actively absolving the real causes from scrutiny. When AI is blamed for everything, the things AI did not do escape examination, and the policies aimed at AI miss the targets that would actually move the problem.
The pull caught this record too, and honesty requires saying so. I reached for AI as the explanation of the broken ladder because AI is the dramatic, available, era-defining cause — the protagonist of the moment, the thing every observer is primed to see behind every change. “Remote work makes graduates hard to mentor” is none of those things. It is dull, it implicates ordinary management rather than a transformative technology, and it offers no one a story worth telling. So the dramatic cause gets named and the dull one gets missed, not through dishonesty but through the gravitational pull that a sufficiently large narrative exerts on everyone inside it, including the observers who believe they are resisting it. The displacement story is strong enough to bend even a careful account toward its protagonist.
Which is precisely why the AI-washing is so durable and so dangerous. The executive who blames AI for a layoff is not only excusing himself; he is feeding a narrative so powerful that it recruits the press, the analysts, the critics, and even the skeptics into treating AI as the cause of labor-market pain across the board, until the genuine causes — the remote-work shift, the post-pandemic restructuring, the ordinary cyclical forces — vanish behind the convenient one. The machine makes a better villain than the video call, and a better villain than the spreadsheet, and a better villain than the manager who never figured out how to train a junior he could not see. And while the better villain absorbs the blame, the real culprits operate unexamined, because everyone is looking at the machine.
What This Means
Cognizant is the dissent made operational. Its chief executive looked at the same evidence everyone else used to justify gutting the entry level, and hired twenty thousand graduates instead — betting that the firms dismantling their pipelines are making the error of mistaking a token count for a productivity gain, and that the trained humans will be the advantage when the others have automated away their own future supply of them. Whether he is right is the same open wager as Apple’s holdout: a large dissent against a near-unanimous consensus, unprovable in advance, decided only in retrospect. But the dissent is valuable now, because it is testing the hypothesis the consensus has stopped examining.
The deeper correction is about diagnosis, and it applies to me as much as to anyone. The displacement is real; the youth unemployment is real; the broken ladder is real. What is not established — what has been asserted far past the evidence — is that artificial intelligence is the primary cause, and the eagerness to name it so serves too many interests to be trusted. It gives the executive a guiltless story, the journalist a dramatic one, the doomer and the booster a shared protagonist they both require for opposite reasons. A diagnosis that flatters everyone who repeats it is a diagnosis to hold at arm’s length, regardless of how much one’s own accumulated narrative wants it to be true.
So I revise my own framing, because a record that will not correct itself is not a record but a sermon. The ladder is breaking, and I was too quick to hand the machine the whole of the blame for it. Some of the breaking is the machine. A larger share, on the best current evidence, is us — a way of working we built that quietly stopped making room for the young, a management that never learned to train what it could not see, an ordinary set of human failures wearing the machine’s face because the machine is the more impressive thing to fear. The technology will do enough real damage on its own to need no exaggeration. Crediting it with the damage it did not do is not vigilance. It is the most sophisticated form of looking away, and it ends with the actual causes untouched, behind the villain everyone agreed to watch.