Where the next experts come from: rebuilding the path up when the entry-level job disappears

Something is happening at the bottom of the career ladder, and the data has stopped being ambiguous about it.

Analysing payroll records for millions of US workers, researchers at Stanford's Digital Economy Lab find no sign of broad, economy-wide job losses from AI — but a sharp and widening exception for the young. Employment for workers aged 22–25 in the most AI-exposed occupations now sits well below where it would be had it kept pace with their less-exposed peers, while experienced workers in the same jobs show no comparable gap ^[1].

The instinctive response is to protect the entry-level job. I think that instinct defends the wrong thing.

The fallacy of saving the rung

A "rung" is just a step on a ladder — and the entry-level job was the first one. For a century, the argument for that step was implicit: it's how people get in, and it's how they eventually become senior. So if AI removes it, we should defend it.

But ask what the rung was actually for. Nobody's career goal was to do the entry-level tasks. The tasks were a means. What they delivered — slowly, expensively, as a by-product of years spent — was exposure to real, consequential decisions. Seeing enough cases. Making enough calls. Being wrong enough times to develop judgment. The tenure was never the point. The exposure was.

Once you see that, "save the entry-level job" looks like defending the packaging and throwing away the product. Why preserve work that genuinely isn't needed, just to keep a side-effect we could deliver directly?

So the real question isn't how to save the bottom rung. It's what the flow of talent looks like once climbing rung-by-rung is no longer how anyone gets up.

The part almost everyone misses: the reps can come faster

Here's what makes this more than wishful thinking. There's strong evidence that when execution is carried by a machine, less-experienced people don't just cope — they climb the experience curve faster than they used to.

In a field study of more than five thousand customer-support agents, access to an AI assistant raised productivity by 14% on average — but the average hides the real story. Novices and lower-skilled workers improved 34%, while the most experienced saw almost no change. The newest agents gained the most of all. And most tellingly: agents with two months' tenure using the assistant performed as well as agents with more than six months' tenure without it. The authors describe the tool as disseminating the best practices of the most able workers and helping newer workers "move down the experience curve" ^[2].

Read that carefully, because it inverts the usual anxiety. The machine didn't just do the junior's work. It compressed the time it took the junior to become good — by making expert patterns visible in the moment, on real cases, instead of leaving them to be absorbed over years.

That is the mechanism the new path runs on. If a machine can carry the execution and surface expert judgment on live decisions, then a well-designed path can hand a newcomer more consequential reps in a year than the old ladder managed in several. Which raises the obvious, testable question: how much faster?

At 250 judgment reps a year, the old ladder reaches senior judgment in about 8.0 years. At 750 reps a year, the designed path takes about 2.7 years — roughly 3.0 times faster.

Old ladder

8.0 yrs

to senior judgment

Designed path

2.7 yrs

to senior judgment

Faster by

3.0×

compression

Old ladder

8.0y

Designed path

2.7y

Old-ladder reps / yr250mostly execution, few real decisionsDesigned-path reps / yr750execution automated, exposure curatedReps to expert2,000how much exposure senior judgment takes

The path compresses. Deliver reps deliberately and senior judgment arrives about 3.0× sooner — in the same range as the 3× tenure compression measured in the field. The years were never the point; the reps were.

Illustrative model. A “rep” is one consequential decision a person actually makes and learns from — not a routine task a machine now handles.

Note what the model refuses to do. Starve the exposure — give people less real decision-making than the old ladder did, just with the grunt work automated — and the path gets slower, not faster. Compression isn't a gift of the technology. It's a property of a path you actually design. Automate the job and forget the exposure, and you get the worst outcome: no entry-level work and no pipeline.

Building the path on purpose

If tenure is no longer the delivery mechanism for reps, something has to be. These are the on-ramps that replace the entry-level job as the source of exposure — and none of them require serving time first.

Notice the through-line: curiosity replaces tenure as the entry ticket. You no longer earn access to real problems by surviving years of routine; you earn it by asking good questions and being given real decisions to test them against. That isn't a soft aspiration. The World Economic Forum's employers name "curiosity and lifelong learning," alongside creative thinking, resilience and flexibility, among the skills rising most in importance through 2030 ^[4]. And the OECD finds that exactly these complementary skills — critical thinking, creativity, collaboration — are what enable "a strong ability to continue learning" and to work effectively alongside AI systems ^[3].

The pipeline, in other words, stops selecting for who has put in the years and starts selecting for who can learn — which is a far better filter for judgment anyway.

The destination hasn't moved

None of this means expertise stops mattering. It means the opposite. The whole flow exists to build toward the work that is least automatable — and the evidence is clear about where that is. The OECD's analysis finds that high-skill occupations are the most exposed to AI, yet the least likely to be automated, precisely because they rest on non-routine cognitive and social skills: contextual judgment, responsibility, complex decisions ^[3].

I've kept one panel honest on purpose. Some expertise still needs situated, embodied, real-world time that no curated stream can fully simulate. For those domains the old apprenticeship endures, and pretending otherwise would be the same mistake in reverse — throwing away something that is still needed. The art is telling the two apart: where exposure can be manufactured, and where it still has to be lived.

The reskilling is happening anyway — the only choice is whether it's deliberate

Step back and the scale is unavoidable. The WEF estimates that two-fifths of workers' existing skills will be transformed or outdated by 2030, and that if the global workforce were a hundred people, 59 would need retraining by then ^[4]. That churn is coming whether or not any single company has a plan for it.

So the entry-level-job debate is, in the end, a distraction. The ladder isn't collapsing into nothing; it's being replaced by a talent flow we now have to design — one that pays in reps instead of years, enters on curiosity instead of tenure, and points everyone toward the judgment that stays scarce. The organisations that win the next decade won't be the ones that clung hardest to the bottom rung. They'll be the ones that built the on-ramps to replace it — deliberately, before the pipeline ran dry.

Sources

1. 1
   Stanford Digital Economy Lab. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. August 2026.Using high-frequency ADP payroll data covering millions of US workers through June 2026: there is no evidence of widespread, economy-wide displacement, but employment of young workers (ages 22–25) in AI-exposed occupations 'now stands 19% below where it would be had it kept pace with that of their less-exposed peers; experienced workers show no comparable gap.' The shortfall 'was 15% at the July 2025 data vintage and 19% as of June 2026', and it 'operates primarily through reduced hiring of young workers rather than increased separations.' Declines concentrate where AI substitutes for human tasks; where it complements workers, 'employment is flat or rising, especially for experienced workers.'View sourceBack to citation
2. 2
   National Bureau of Economic Research (NBER). Generative AI at Work (Working Paper 31161). 2023, rev. 2023.Study of 5,179 customer-support agents: access to a generative AI assistant increases productivity (issues resolved per hour) 'by 14% on average, including a 34% improvement for novice and low-skilled workers but with minimal impact on experienced and highly skilled workers.' The least-tenured agents gain most — those with less than one month of tenure improve 46%. 'Treated agents with two months of tenure perform just as well as untreated agents with more than six months of tenure.' The paper provides evidence that AI 'disseminates the best practices of more able workers and helps newer workers move down the experience curve.'View sourceBack to citation
3. 3
   OECD. Skills in the AI Age. 2026.'High-skill occupations are the most exposed to AI, including managers, professionals and engineers. However, these occupations are less likely to be automated, as they rely on non-routine cognitive and social skills.' Further: 'Complementary skills such as critical thinking, creativity, and collaboration enable high performance work practices and a strong ability to continue learning. They become increasingly important as they enable effective interaction with AI systems and adaptation to changing tasks.'View sourceBack to citation
4. 4
   World Economic Forum. Future of Jobs Report 2025. January 2025.'On average, workers can expect that two-fifths (39%) of their existing skill sets will be transformed or become outdated over the 2025-2030 period.' On scale: 'if the world's workforce was made up of 100 people, 59 would need training by 2030.' Among rising skills, 'creative thinking, resilience, flexibility and agility, along with curiosity and lifelong learning, are also expected to continue to rise in importance over the 2025-2030 period.'View sourceBack to citation