The Vanishing Bottom Rung: AI Pulled the Learning Out of Junior Work

On June 15, PwC (one of the Big Four accounting firms) published a report that scanned more than a billion job ads across 27 countries. One number in it held my eye for a long time: entry-level roles touched by AI are seven times more likely to demand "senior skills" (judgement, leadership, the things we used to ask only of mid- and upper-level staff) than other entry-level roles.

Put plainly, a lot of jobs still labelled "junior" or "open to new grads" have quietly raised the bar to a senior height. In the same report, these redefined entry-level roles have grown 35% since 2019, while traditional entry-level jobs untouched by AI have shrunk by 10%.

When most people hear "AI hits jobs," they reach for "AI is taking jobs." That frame is too blunt. What's actually happening is more precise and more troubling: what AI removed is the learning buried inside junior work. The shell of the job is still there; the apprenticeship at its core is gone.

Let me unpack that. What did a new hire used to do? Compile materials, run the first round of research, format documents, write up meeting notes, fold scattered bits into a single page. At the time these looked like grunt work, but they had a second identity: they were the apprenticeship. While you did these unglamorous things, you were quietly absorbing how the industry runs, what good work looks like, what the boss cares about. The pay was a salary; the bonus was experience.

These are exactly the tasks AI now does most smoothly and most cheaply. First-pass research, first drafts, summaries, formatting, a model spits out a version in seconds at a cost near zero. So a company does the obvious math: if AI can do the grunt work, why pay a slow, error-prone newcomer to do it?

On paper the math checks out. The problem lives where the paper can't see.

That grunt work you cut was also the single channel through which this company manufactured its own future senior staff. Close the channel and you save one junior salary today, but the senior employee you'll want in five years has nowhere to grow from. PwC's Global Workforce Leader, Pete Brown, put it carefully: "AI is removing some of the routine work that once acted as an apprenticeship, while increasing demand for judgement, leadership and adaptability much earlier in careers." In plainer words: you used to have five years to climb; now the company wants you to show up on day one with the judgement of someone five years in.

This lands as a hard problem for two kinds of people.

For owners and managers, it's a hidden long-term risk. The productivity is real. In the report's most AI-exposed sectors, output per worker rose 34% over seven years, and the top fifth of companies saw 163%. Against numbers like that, cutting junior roles and pairing senior people with AI is almost the inevitable short-term move. Just be clear about the trade: you're spending five years of future talent inventory to buy this year's margin. When the whole industry does it at once, in a few years the market will be short an entire generation of mid-career talent, and senior salaries will climb high enough to make you regret the pennies you saved today. The wage premium for AI skills is already 62%, up from 57% last year, and still rising.

For anyone just starting out, or with a kid who just graduated, the old path needs a rethink. "Start at the bottom, do the grunt work well, climb slowly," advice repeated for decades, just lost a chunk of its foundation. Grunt work is no longer your apprenticeship, because the grunt work has been outsourced to AI. What you can trade for a seat, from day one, has to be the part AI still can't do well: judging which result to trust, breaking a vague problem into clear pieces, making the call when there's no right answer, getting a group of people pointed the same way. These used to be grown by putting in the years. Now you have to find a way to practise them early.

Running a swarm of AI agents (programs that read material, call tools, and run tasks on their own) solo every day, I feel this directly. The models take all the junior work off my plate, and the time I save doesn't make my days lighter; it forces every saved hour onto judgement: is this worth building, can I trust this result, should I let it run this step on its own. One person out-producing a small team comes down to exactly that. AI drove the price of grunt work to the floor, and the only thing still worth paying for is your capacity to set direction and own the outcome.

So I won't read this report as bad news. It just states one thing very clearly: the era when a position waited around for you to mature is over. Skill can still grow gradually, but the part of it the market is willing to pay for now points, earlier and more directly than ever, at judgement itself.