AI is taking the work that used to train your juniors

For decades, most firms grew expertise the same way. New people did the routine work: the first-pass research, the basic reconciliation, the document review, the early drafts. They were slow and they made mistakes, and that was the point. Doing the work, and having it corrected, was how they learned to judge it. That bottom rung of work is exactly what AI now does well and cheaply. PwC's 2026 Global AI Jobs Barometer, drawn from more than a billion job adverts, found that the entry-level roles most exposed to AI are now seven times more likely to demand traditionally senior skills like judgement and leadership. So a question lands on every leader who runs a team: if AI does the work that used to train your people, how will they learn judgement instead? Leave it to chance, and the apprenticeship quietly disappears.

What the figures actually say

PwC analysed entry-level hiring and found two things moving at once. The early-career jobs most exposed to AI grew 35% between 2019 and the present, while other entry-level roles shrank 10%. The surviving junior jobs are not the old ones with a chatbot bolted on. They are different jobs, asking earlier for skills that used to take years to build. Based on 2.4 million entry-level postings in the United States, those AI-exposed roles are seven times more likely to call for senior, human-intensive skills than the least exposed ones.

That points at a real change, but a single report is thin evidence on its own, so it is worth pairing with harder data. The Stanford Digital Economy Lab, using payroll records from the United States, found that workers aged 22 to 25 in the most AI-exposed jobs saw a roughly 13% relative decline in employment since late 2022, while older colleagues in the same roles kept growing. The drop was concentrated where AI does the task outright rather than helping a person do it. The researchers were careful about what this proves: it is early evidence, not a settled law. Both are US figures, and the UK labour market is its own thing. But the direction is consistent, and it matches what a lot of leaders are already seeing in their own hiring.

The headline that this is about redundancies misses the quieter mechanism. Firms are mostly not making juniors redundant. They are not replacing them when they leave, because the work a junior used to cut their teeth on is now done by a tool. The cost of that does not show up this quarter. It shows up in about five years, when the senior people who would have come up through that rung are not there, because the rung was removed before they could climb it.

Why this lands hardest on professional services

The firms most exposed are the ones built most tightly around the apprenticeship. An accountancy practice taught reconciliations and basic preparation by handing them to juniors. A law firm taught judgement through document review and first-draft research. A corporate finance team taught a graduate the business by making them build the model and check the numbers. A recruiter learned the market by working through long lists of candidates by hand. In each case the routine work was never only about output. It was the training ground, and the judgement came as a by-product of grinding through it.

AI is good at precisely that layer. It drafts, summarises, reconciles, reviews and ranks. Used well, that is a real gain: the work gets done faster and people are freed from drudgery. The trouble is that the drudgery was doing a second job nobody had written down. Take it away without replacing it, and you keep the output while losing the apprenticeship. We can call this the apprenticeship gap: the distance between work an AI can now do and the judgement a person used to build by doing it.

This is not an argument for keeping busywork to toughen people up. It is an argument for being honest that the old training was a side effect, and that side effect now has to be designed on purpose rather than assumed.

The people on the other side of it

It helps to look at this from where a junior is standing, not only from the firm's side of the ledger. The ladder they expected to climb is having its bottom rungs pulled up. There are fewer entry roles in exposed fields, and the ones that remain ask, on day one, for the judgement and the client presence that used to be earned over years. That is a harder, narrower door than the one their managers walked through. A leader who treats this purely as a cost saving, and not also as a duty to the people they hire, will tend to hollow out their own future bench without noticing, and leave a cohort of capable people with nowhere to start.

What a leader should actually do

The useful response is not to ban the tools or to slow the work down. It is to take the training that used to happen by accident and make it deliberate.

Start by deciding what juniors should still do by hand, and why. Some routine work is worth keeping not because a person is faster at it, but because doing it builds the judgement the role needs later. That is a different test from pure efficiency, and it is one only a leader can set. It sits next to the related call about what should never be handed to AI at all, the decisions that must stay with a person who can answer for them.

Then turn AI itself into the teaching tool. The most valuable thing a junior can now do is check, correct and improve AI output, because that is where judgement forms when the first draft is no longer theirs. A graduate who reviews an AI-drafted analysis and has to find what is wrong with it is learning the same discernment the old first-draft work taught, faster and on harder examples. That only works if a senior person reviews the review, and if being wrong is treated as part of learning rather than a failure. The apprenticeship moves up a level, from producing the work to judging it, but it still has to be run.

None of this happens on its own, and that is the leadership part. Microsoft's 2026 Work Trend Index found that only about one in four workers say their leadership is clearly aligned on AI. A team will not rebuild how it develops people while the people at the top are still undecided about what AI is for. This is the same pattern we see across adoption generally: the gap is rarely the staff, it is the direction set above them.

This is the kind of question the Good Transformer Lessons for Leaders sessions are built around: not which tool to buy, but how your team actually works once AI is doing the routine layer, and where to keep human judgement growing rather than letting it thin out. If that is worth an hour for your firm, book a discovery call.

The decision in front of you is concrete, and it is yours to make this year. Look at one team and ask where its judgement used to come from. If the answer is "the work AI now does," then you have a development gap to fill on purpose, before the bench you will need in five years quietly fails to form.

FAQ

Is AI actually replacing entry-level jobs?

The evidence points to pressure on the junior rung rather than a clean replacement. PwC's 2026 Global AI Jobs Barometer found entry-level roles most exposed to AI are seven times more likely to require senior skills, and Stanford's analysis of US payroll data found workers aged 22 to 25 in the most AI-exposed jobs saw a roughly 13% relative decline in employment since late 2022. In many firms the mechanism is not redundancy but quietly not replacing juniors when they leave, because the routine work they used to do is now automated.

Why does it matter if AI does the routine junior work?

Because that work did two jobs. It produced output, and it trained judgement as a by-product. Reconciliations, document review, first-draft research and candidate sifting were how people learned to tell good from bad in their field. Remove the work without replacing the training, and you keep the short-term output while losing the way your future senior people were made.

What should leaders do to keep developing people?

Decide deliberately what juniors should still do by hand because it builds judgement, not just because it is efficient. Then use AI as a teaching tool: have early-career people check, correct and improve AI output under senior review, since judging a draft is where discernment now forms. Treat being wrong as part of learning, and make sure someone experienced reviews the review.

Does this apply in the UK, given the data is from the US?

The two studies cited use US data, and the UK labour market has its own dynamics, so the exact figures should not be read across directly. The structural point is not US-specific: any firm that trained people by handing them routine work AI can now do faces the same development gap. Treat the numbers as a signal of direction, and look at your own hiring and your own teams for the local picture.