AI Arrived Before Apprenticeship

For a long time, becoming good at something involved doing a lot of work that you would eventually stop doing.

The junior lawyer reviewed documents.

The new analyst cleaned spreadsheets.

The young designer produced rough variations.

The first-year researcher summarised papers.

The apprentice programmer fixed simple bugs.

The intern took notes.

Much of this work was tedious.

It was also how people learned.

Generative AI has arrived with a compelling proposition:

perhaps humans no longer need to do so much tedious work.

That sounds wonderful until we ask a slightly different question.

What if the tedious work was also the apprenticeship?

This is one of the least discussed problems in the conversation about AI and jobs.

We spend enormous amounts of time debating whether artificial intelligence will replace professions.

The more immediate problem may be what happens to the bottom rung of those professions.

A 2026 World Economic Forum report produced with PwC estimated that more than one in three young workers globally are in occupations with medium-to-high exposure to AI-driven task change.¹

The earliest labour-market evidence deserves caution, particularly because much of it comes from wealthy countries.

But it is beginning to show why entry-level work deserves special attention.

Researchers at Stanford's Digital Economy Lab, analysing payroll records covering millions of US workers through June 2026, found no evidence of widespread economy-wide job displacement associated with AI. Yet employment among workers aged 22 to 25 in highly AI-exposed occupations stood 19 per cent below where it would have been had it kept pace with their less-exposed peers. Experienced workers in the same occupations did not show the same gap.²

It is one country and an evolving working paper, not proof of a global future.

But it is precisely the kind of asymmetry worth watching.

This creates a strange possibility.

AI may not eliminate the architect, accountant, programmer, designer or lawyer.

It may instead eliminate enough of the work through which someone became one.

There is already a version of this problem in education.

A student who cannot yet write a strong essay can now produce one.

Someone who cannot program can generate functioning code.

Someone who cannot draw can produce an image.

Someone who has never synthesised a complicated body of research can ask a machine to do it.

We normally describe this as cheating when it happens in school and productivity when it happens at work.

That distinction will become harder to maintain.

The more important distinction may be between using a tool to extend competence and using a tool to bypass the formation of competence.

Experienced professionals often know which parts of an AI-generated output are wrong because they possess the expertise the machine is augmenting.

Beginners may not.

Here is the paradox of powerful tools for beginners.

The people most capable of getting value from them may be people who already learned without them.

None of this is an argument for making young people perform pointless tasks because previous generations had to.

Technology has always removed work.

Calculators changed mathematics.

Search engines changed research.

Computer-aided design changed engineering.

Nobody seriously proposes that young architects should draw every technical plan by hand simply to preserve tradition.

The challenge is identifying what the obsolete labour was secretly teaching.

Perhaps document review taught lawyers to recognise patterns.

Perhaps manually debugging code exposed programmers to how systems fail.

Perhaps writing ten bad drafts gave a writer an internal sense of a sentence.

Perhaps performing basic analyses taught a researcher to notice when numbers did not make sense.

If AI produces the output and removes those encounters, we need another way to produce the learning.

Schools face exactly the same problem.

Employers may have to confront it even faster.

For decades, organisations have complained that graduates arrive without experience while relying on junior positions to create experienced workers.

If entry-level tasks become easily automated, companies could discover that they have optimised away their own talent pipeline.

The answer is probably not protecting every entry-level task.

It is designing better apprenticeships.

A junior employee using AI might produce work faster, then spend more time explaining why the output is correct.

They might compare machine-generated alternatives, diagnose errors, observe experienced colleagues making decisions or take responsibility for increasingly ambiguous problems.

In other words, the unit of learning shifts from production toward judgment.

This could ultimately produce better early careers.

Nobody needs nostalgia for photocopying documents or formatting PowerPoint slides at midnight.

But it will not happen automatically.

If the economic incentive is simply to replace five junior workers with one experienced worker plus AI, organisations may do exactly that.

The benefits will appear immediately.

The cost may only become visible five years later when there are fewer experienced people to hire.

Young people are entering the workforce at a peculiar moment.

For previous generations, competence generally came before leverage.

Today's beginner can acquire enormous leverage immediately.

We will have to figure out how to preserve the difficult, slow and sometimes frustrating process through which judgment forms without pretending that the tools do not exist.

The question is no longer whether young people should use AI.

They will.

The harder question is:

How do you learn to do something when a machine can do it before you can?