If AI does the work we used to learn from, how will we build tomorrow’s experts?

Think back to the first few years of your working life.

Drafting something for the first time. Looking for information. Organising data. Fixing a presentation. Checking a document. Preparing a summary that somebody more experienced would later review.

A lot of that work was slow. Some of it was frankly quite dull.

Today, AI can do part of it in seconds.

In many cases, that is progress.

But before we remove all of that work, there is another question worth asking: was it simply low-value work, or was some of it also how we learned?

Because there was more than wasted time inside an imperfect first draft.

Somebody was trying. Somebody was getting things wrong. Somebody was receiving corrections and gradually beginning to understand why one answer worked better than another.

This is where the conversation about AI and entry-level jobs becomes more interesting than the familiar question of how many roles artificial intelligence might replace.

For organisations, the harder question may be this:

if we automate part of the work through which people used to learn, where will tomorrow’s experience come from?

We do not need to preserve boring work

There is no good reason to keep repetitive tasks just because somebody learned from doing them twenty years ago.

AI can save time, accelerate research, produce first drafts, support analysis and remove a great deal of administrative effort.

The point is not to protect the past.

It is to understand what else might disappear with the task.

A task can be automatable and still have served a learning purpose.

Imagine someone preparing a client analysis for the first time. The real learning may not come from the hours spent gathering information. It may come from deciding what matters, spotting a contradiction, forming a hypothesis and then testing it with someone more experienced.

If AI produces an excellent summary straight away, the organisation has gained efficiency.

It has not necessarily built capability.

Those are two different outcomes.

The first rung may not be disappearing. It is changing

This is an area where easy conclusions are tempting.

The evidence available today does not show that AI, on its own, is responsible for weaker employment outcomes among young people. The OECD points out that some of the deterioration in early-career labour-market outcomes was visible before the widespread adoption of large language models. Eurofound reaches a similarly cautious conclusion for Europe.

That does not mean AI is leaving entry-level work unchanged.

It is changing what organisations expect from people when they enter.

PwC’s 2026 Global AI Jobs Barometer finds that entry-level roles with greater exposure to AI are increasingly asking for capabilities previously associated with more experienced professionals: judgement, leadership, creativity and human interaction. In its US analysis of 2.4 million entry-level job postings, those AI-exposed roles were seven times more likely to require skills historically associated with senior positions.

A US study by Strada, based on nearly 1,500 executives and talent leaders, points in a similar direction. 42% of respondents said AI had increased the analytical and judgement-related responsibilities given to entry-level employees, while 41% reported a reduction in foundational or skill-building tasks.

That creates a paradox.

We may start asking for judgement earlier while removing some of the experiences through which judgement used to develop.

Entry-level does not mean young

There is another assumption worth questioning.

When we talk about an entry-level employee, it is easy to picture a 22-year-old starting their first job.

But somebody can be highly experienced and still become a beginner again.

They may change industry, move into a new profession, take on their first management role or encounter a technology that changes how their expertise is applied.

Experience is not only about age.

It is contextual.

You can know a great deal in one field and still be a beginner when facing a new technology, market or type of problem.

This distinction matters increasingly in organisations where tools and capabilities are changing quickly.

It is also consistent with a principle at the heart of multigenerational management: age, seniority, experience and career stage are related, but they are not the same thing.

AI can accelerate output. It cannot shortcut experience

Someone can use AI to produce a strong presentation without yet knowing how to recognise a strategically weak one.

They can receive a convincing answer without knowing what information is missing.

They can generate code, an analysis, an email or a commercial proposal of impressive quality without yet being able to explain why that solution is appropriate in that particular context.

That is not automatically a problem.

It becomes a problem when we confuse the quality of the output with professional maturity.

Experience is also built when something fails.

It develops when a person has to defend an assumption in front of somebody more knowledgeable, when a decision creates an unexpected consequence, or when two situations that looked identical turn out to require different responses.

AI can shorten the route to an answer dramatically.

But professional experience is not simply an accumulation of correct answers.

It also includes knowing when the answer in front of you is not enough.

An organisation can become highly efficient and surprisingly bad at teaching

At this point, the issue stops being about individual junior employees.

It becomes an organisational question.

Imagine an organisation that automates more and more straightforward tasks while managers and senior colleagues simultaneously spend less time explaining, observing and giving feedback.

In the short term, the result may look excellent.

More output. Less time.

But efficiency and capability-building are not the same thing.

If the learning that once happened inside the work disappears, organisations need to decide where that learning will happen instead.

Otherwise, they may discover later that they have become very good at producing work and much less good at producing people who can handle complex work independently.

The question therefore cannot only be:

“Where can we find people who are already prepared?”

It also has to become:

“How good are we at developing the people who join us?”

That is a more uncomfortable question because it puts part of the responsibility back on the organisation.

Managers are becoming part of the learning infrastructure

For a long time, a significant amount of workplace learning happened almost accidentally.

You sat next to somebody with more experience. You watched. You asked questions. You prepared something, received a page full of corrections and did it slightly better the next time.

AI changes part of that process.

If the first attempt is increasingly produced by a machine, managers cannot limit themselves to reviewing the final output.

They need to help people build the thinking behind it.

That can mean asking:

“Why did you choose this approach?”

“What assumption are you making?”

“What do we still not know?”

“Which part of the AI-generated answer did you verify?”

“What would change if the context were different?”

This is closer to the role of a manager who coaches than to that of a manager whose main job is checking whether a task has been completed.

It is part of the broader change already taking place in the work of managers in the AI era.

And it is one reason why management training becomes more, rather than less, important as AI spreads through organisations.

AI literacy is necessary. It is not enough

Many organisations are rightly investing in AI literacy.

They should.

But knowing how to use a tool does not automatically solve the learning problem.

Someone can become highly skilled at prompting while remaining weak at judging the quality of an answer.

They can produce five alternatives in seconds without yet having strong criteria for choosing between them.

They can accelerate a process without understanding which part of that process should not be automated.

The more useful question may therefore be moving from:

“Can you use AI?”

to:

“Can you work effectively with AI without outsourcing the development of your own judgement?”

Five things HR and managers can start redesigning

Organisations do not need to slow down AI adoption. They need to design more deliberately around it.

• Separate repetitive work from developmental work. Automating a task that adds little value is efficiency. Automating a task that also provided learning without replacing that learning creates a gap.
• Ask for the reasoning, not only the result. A strong output can hide fragile understanding. Asking people to explain assumptions, criteria and alternatives makes learning visible.
• Delegate decisions progressively, not only tasks. Autonomy does not appear because somebody is suddenly given harder work. It grows through increasing responsibility, feedback and room to correct mistakes.
• Protect time for mentoring and feedback. If people development is something managers do only when there is time left over, operational urgency will always win.
• Let knowledge move in both directions. More experienced colleagues can provide context and judgement. People who work more naturally with emerging tools may contribute practices and perspectives others have not yet developed. This is where mentoring and reverse mentoring can become part of the work rather than symbolic initiatives.

The problem is not that junior employees are using too much AI

It is easy to turn this conversation into a generational complaint.

“We actually had to learn how to do the work.”

“Younger employees cannot do anything without ChatGPT.”

That would be a convenient shortcut, and not a particularly useful one.

If a tool enables people to work faster and better, using it is a rational choice.

The responsibility of an organisation is not to recreate the inconvenience of the past.

It is to ask a different question:

what capability used to develop inside that effort, and how will we develop it now?

That keeps a technological shift from becoming a judgement about people.

Senior professionals will have to learn differently too

This is not only an early-career issue.

As technologies, processes and roles change, someone with twenty years of experience can find that one part of their expertise remains extremely valuable while another has to be rebuilt.

That is not an age issue.

It is what happens when somebody encounters a problem for which their existing experience is not yet enough.

Learning inside organisations may therefore become less linear.

We will not simply have experienced people teaching inexperienced people.

Depending on the problem, the same person may be an expert in one conversation and a beginner in the next.

That makes knowledge transfer, reciprocal mentoring and collaboration across different levels of seniority increasingly important.

Experience will continue to matter.

But it will need to move.

Building tomorrow’s experts is an organisational choice

AI may free us from many of the tasks through which previous generations learned.

That can be very good news.

But removing simple work does not automatically create experienced people.

Judgement, autonomy, contextual understanding, decision quality and relational maturity still require feedback, mistakes, increasing responsibility and exposure to people who know more than we do.

Some of the experiences that used to emerge almost naturally from everyday work may now need to be designed much more intentionally.

That makes this a strategic question for HR and Learning & Development.

The question is not only:

what skills will we need five years from now?

It is also:

what kind of managers will we need to develop them?

Technology can be purchased quickly.

Building people who know how to use it with judgement takes something different.

It takes time, culture and managers who know how to develop other people.