AI may not need to eliminate entire professions to reshape the labour market. Automating enough entry-level work could change how people acquire the experience required to become experts.
Much of the debate about artificial intelligence and employment revolves around a binary question:
Will AI replace jobs?
So far, the evidence doesn’t support a simple answer.
AI adoption has accelerated, and employers increasingly expect it to affect their workforce. Yet large-scale displacement has not appeared across aggregate employment data. The effects that are visible are uneven, concentrated in particular occupations, tasks and groups of workers.
One pattern deserves particular attention.
Young workers in some highly AI-exposed occupations appear to be doing worse than more experienced workers in the same labour market.
That raises a different question from whether AI will eliminate professions.
What happens if it automates some of the work through which people learn to become professionals?
The bottom rung matters
Professional expertise rarely appears fully formed.
A junior analyst reviews documents, cleans data and prepares first drafts. A young lawyer researches cases and works through contracts. A developer fixes relatively simple problems before designing complex systems. A new accountant reconciles accounts before advising organizations on difficult financial decisions.
Some of this work is repetitive.
That is precisely why it is attractive to automate.
But junior work can perform two functions simultaneously.
It produces an output for the organization.
It also produces a more capable worker.
The first function is easy to observe. The second is slower and harder to measure.
An organization deciding whether to automate a task therefore faces an asymmetry. The immediate productivity gain may be visible within weeks. The loss of a learning pathway may not become visible for years.
An early labour-market signal
Recent evidence makes this more than a hypothetical concern, although causality remains uncertain.
Stanford’s 2026 AI Index reports that AI-related labour-market effects are appearing unevenly and are particularly visible in hiring pipelines and among younger workers in exposed occupations. Employment among software developers aged 22 to 25, for example, fell nearly 20 percent from 2024.
More detailed work from Stanford’s Digital Economy Lab finds a pronounced divergence among early-career workers aged 22 to 25. Employment has weakened in occupations with greater AI exposure, while less-exposed occupations have generally performed better. The relationship is especially noticeable where AI usage is oriented toward automation rather than augmentation.
This is suggestive, not conclusive.
Exposure to AI is not the same as displacement by AI. Technology hiring has been affected by interest rates, post-pandemic adjustments and other economic forces. Occupational categories are broad. Labour markets can also adapt in ways that aren’t immediately visible in employment statistics.
The evidence therefore doesn’t justify the claim that AI is already eliminating entry-level careers.
It does justify watching them closely.
The experience paradox
Suppose AI becomes particularly effective at work normally assigned to junior employees.
For an individual organization, using it can be entirely rational.
A senior professional supported by AI may produce work that previously required several junior colleagues. Costs fall. Output increases. Experienced workers spend less time reviewing routine material.
But if many organizations make the same decision, an unexpected problem can emerge.
There may be fewer places for inexperienced workers to become experienced workers.
This is an experience paradox.
The tasks easiest to automate may also be tasks through which tacit knowledge, professional judgment and contextual understanding are acquired.
Reading hundreds of contracts isn’t merely a way to produce contract summaries.
Debugging ordinary code isn’t merely a way to fix ordinary bugs.
Reviewing routine cases isn’t merely a way to clear a queue.
Repeated exposure to relatively simple situations can build the pattern recognition needed to handle difficult ones later.
The value of junior work may therefore be systematically underestimated if organizations measure only its immediate output.
AI may change the shape of a career
This doesn’t necessarily imply fewer experts in the future.
AI could change professional development rather than destroy it.
Junior workers may be able to tackle more sophisticated problems earlier because AI handles routine components of their work. Training could become faster and more individualized. Organizations might deliberately redesign apprenticeships around simulation, supervision and accelerated exposure to difficult cases.
There is evidence supporting a more optimistic interpretation.
Stanford research published in July 2026 argues that AI could simplify some tasks enough to allow less-skilled workers to enter higher-paying occupations, potentially expanding opportunity rather than restricting it.
Anthropic’s June 2026 Economic Index survey also complicates a straightforward displacement narrative. Early-career respondents reported greater concern about job loss and said AI could perform a larger share of their work, but users who delegated more work to AI were, on average, more optimistic about their own labour-market prospects. Anthropic’s sample consists of Claude users and is not representative of the entire workforce, so the finding should be interpreted cautiously.
AI could therefore produce two very different career structures.
In one, automation removes junior tasks and narrows entry routes into professions.
In the other, AI becomes a form of cognitive leverage that allows inexperienced workers to progress faster.
Both are plausible.
The difference may depend less on the technology than on how organizations redesign work around it.
Productivity has a time horizon
This creates a measurement problem.
Imagine that replacing some junior work with AI increases productivity this year.
From the perspective of the organization, that is a real gain.
But suppose the same change reduces the number of people acquiring the experience necessary to perform senior roles five or ten years later.
That cost doesn’t appear in today’s productivity calculation.
Organizations therefore may optimize individually in ways that weaken a capability collectively.
The problem is familiar in other forms.
Firms can reduce training because trained workers might leave. Industries can underinvest in skills because the benefits are shared across employers. Public institutions can cut analytical capacity because expertise appears expensive until a crisis reveals its value.
AI could intensify this dynamic by making the immediate substitute for junior labour unusually attractive.
The relevant economic question is therefore not only whether AI raises output per worker.
It is whether it changes the process through which productive capability itself is formed.
Expertise is more than task completion
This distinction also challenges a common way of estimating AI exposure.
Occupations are often decomposed into tasks, and researchers estimate which tasks AI can perform.
That is useful.
But a profession isn’t simply the sum of its current tasks.
Some tasks have developmental value.
Others create relationships, institutional memory or contextual understanding. Some apparently inefficient activities expose junior workers to enough variation that they eventually develop judgment.
If those functions matter, replacing a task and replacing its role in capability formation are not equivalent decisions.
This resembles a broader distinction between AI adoption and AI capability.
An organization can automate more work without necessarily becoming more capable over the long term.
The immediate output may improve while the system that creates future expertise becomes weaker.
What would weaken this argument?
Several developments could make the concern much less important.
AI could itself become an effective training environment, giving junior workers exposure to more cases and better feedback than traditional apprenticeship models.
Organizations could redesign professional development so that experience is accumulated through different tasks.
Entry-level workers could become substantially more productive without reducing their numbers, allowing organizations to expand output instead of shrinking hiring.
New occupations and career pathways could also emerge faster than old ones contract.
And the current weakness in employment among young workers in AI-exposed occupations may ultimately prove to have causes largely unrelated to AI.
The career-ladder hypothesis should therefore be treated as a signal to investigate, not a forecast.
Watch next
Don’t watch only unemployment.
Watch entry-level hiring.
Compare employment trends for younger and older workers within the same AI-exposed occupations. Watch whether junior-to-senior ratios change inside professional firms. Look for declining graduate recruitment alongside stable demand for experienced workers.
Watch what happens to apprenticeships, internships and professional training.
And pay particular attention to organizations that redesign junior roles rather than simply removing them.
If AI changes the labour market first through career entry rather than mass unemployment, aggregate employment statistics may detect the transition relatively late.
By then, the more important change may already have happened.
The question will no longer be simply whether AI can perform professional work.
It will be whether institutions have preserved a way for people to learn how to do the work that remains.
Leave a Reply