There is an old bargain at the bottom of almost every trade. You show up knowing nothing, you do the dull, repetitive, low-stakes work that nobody senior wants, you do it badly and then less badly, and somewhere in the doing you pick up the thing you were actually there to learn. The filing, the note-taking, the first drafts nobody read, the tickets nobody wanted. It was tedious and often underpaid and we called it an apprenticeship, because it was one. It was also the on-ramp: the slow lane where you merged into the career before anyone expected you to drive.
PwC has now put a number on what happens when you take that lane away. Its 2026 Global AI Jobs Barometer reads more than a billion job ads across 27 countries. Inside it, a US cut of 2.4 million entry-level postings finds the roles most exposed to AI seven times more likely to demand skills that used to arrive much later: judgment, leadership, stakeholder management, strategic decision-making. The firm has a word for it. Seniorization. In the most exposed fields, 52 percent of the new skills showing up in junior postings were skills once written for experienced hires. In the least exposed fields, it was 7 percent. That gap is the whole story.
The mechanism is not mysterious, and PwC does not pretend it is. "AI is removing some of the routine work that once acted as an apprenticeship," says Pete Brown, the firm's global workforce leader, in the report's own release. Read that twice. The routine work was not waste. It was the apprenticeship. The software came for exactly the tasks a newcomer used to cut their teeth on. AI, as ADP's chief economist Nela Richardson told Fortune, absorbs tasks before it absorbs jobs, and it reaches first for the retrievable, the summarizable, the schedulable, the junior work. What is left for the newcomer is the part you were supposed to grow into. The floor and the ceiling have been welded together. You are asked to enter as a junior and perform as someone who has already put in the years.
The honest version is more interesting than the panic. Entry-level work is not vanishing. PwC counted about 11 million early-career postings in 2025, up from 7.3 million in 2018. Openings for these seniorized junior roles grew 35 percent since 2019 while the plainer entry-level jobs shrank 10 percent. The market did not shed the beginner. It repriced them.
And repricing is the part that should hold our attention, because judgment is not a file you download. Stakeholder management is not a weekend course. These are the things you accumulate by being paid to stand near power for a few years, watching how the room works, making small decisions that don't matter until they do. PwC does gesture wider than the worker. Its US chief AI officer, Dan Priest, says that "employers, educators, and policymakers all have a role to play in helping people build those capabilities earlier." Fair enough. But the instruction that actually reaches the twenty-three-year-old, in Fortune's summary of the firm's guidance, is blunter: develop senior skills fast. And that is the sleight of hand, the oldest one in the genre: take a change made in boardrooms and hand most of it back to the individual as homework. Be more, sooner, on your own time. The structural change gets a personal to-do list.
Ask the three questions I always ask. Who benefits, who carries the risk, who gets to leave. The firms deploying AI most heavily are, by PwC's own count, growing headcount faster than their peers, 52 percent against 36 percent since 2018, and paying an AI-skills premium of 62 percent. The gains are concentrating: the top fifth of the most exposed companies posted 163 percent productivity growth. So the risk lands, as it usually does, on the person with the least cushion. The squeeze is already visible in the graduate numbers: in the last quarter of 2025, recent graduates were unemployed at 5.7 percent, above the national rate, and 42.5 percent were underemployed, in jobs that did not need their degree. A crunch like that does not fall evenly. If the entry ticket is now experience you can only buy with prior experience, then the people who clear the bar tend to be the ones who could already afford proximity, the unpaid internship, the family friend's introduction, the school that teaches you to speak like a manager before you have managed anything. Seniorization does not abolish the class filter on good work. It moves it earlier and makes it heavier.
None of this is the machine's fault. There is a version of this shift that is genuinely good, where a twenty-three-year-old is spared two years of soul-flattening drudgery and gets to do interesting work. PwC leans on that version, and so does the more cautious data. Stanford's Erik Brynjolfsson, whose Canaries dashboard tracks payroll for 4.6 million workers, finds employment for 22-to-25-year-olds in the most AI-exposed jobs shrinking about 3.8 percent a year and not bouncing back, as Fortune reported. "Whatever it is, it's not going away," he says, in that same account. The MIT economist Daron Acemoglu, a skeptic of the productivity story, puts the real gains far below the boardroom figures, per Fortune. So the size of all this is a genuine fight. The direction is not.
So the apprenticeship is being unbundled, and nobody has said who now owns the parts. The employer got the productivity. The AI got the grunt work. The young worker got a job description that assumes a decade they have not lived yet. The rungs did not fall off the ladder. They were lifted up out of reach and we were told to jump higher. The obvious question, the one PwC's report circles but will not land on, is who is supposed to pay for the learning now that the paid version has been automated away. Because somebody used to. It was the job.
