The first thing Paul noticed on his first day was the silence. Not the silence of an empty office. The place was full. Screens glowed. Meetings happened. Deadlines moved across digital dashboards. Work was being done everywhere. Yet something seemed missing.

A decade earlier, someone in his position would have spent weeks preparing reports, cleaning spreadsheets, gathering information from scattered sources and building presentations that senior managers would eventually critique. It was tedious work. Often repetitive. Occasionally frustrating. Now, most of it happened in minutes.

The team used AI systems to generate summaries, analyse documents, prepare slides and draft recommendations. Arjun’s role was not to produce the first version. His role was to evaluate it. The difficulty was that he did not yet know how. 

He had arrived at the workplace carrying excellent academic credentials. He knew how to submit assignments, clear examinations and navigate software. What he lacked was the instinct that comes only from wrestling with a problem long enough to understand where mistakes hide. 

Nobody had intentionally removed his training ground. It had simply disappeared.

The public conversation about artificial intelligence is often framed around jobs. Which professions will survive? Which tasks will be automated? Which industries will change first? These are important questions. Yet beneath them lies another question that may prove more consequential. If machines increasingly perform the beginner’s work, where will future expertise come from?

For centuries, societies have relied on a remarkably simple mechanism for developing competence. Beginners started at the bottom. The first rung of almost every profession involved repetitive tasks. Junior lawyers reviewed documents. Young journalists verified facts. Apprentice carpenters measured wood. Medical residents observed senior doctors. Entry-level analysts assembled data. Graduate trainees sat through meetings and took notes. The work itself was rarely glamorous. But its purpose was something deeper. Repetition trained attention, mistakes produced feedback, proximity allowed observation and slowly, often invisibly, judgment emerged. What appeared to be labour was also education.

The sociologist Richard Sennett once wrote about craftsmanship as a process through which people learn to do something well for its own sake. Such learning is difficult to compress because it depends on experience accumulating over time. Expertise is not merely knowledge stored in the mind. It is pattern recognition shaped by countless encounters with reality. A lawyer develops an instinct for weak arguments because she has encountered many. A teacher learns to recognise confusion in a student’s face because he has seen it repeatedly. A manager learns which numbers matter because he has spent years watching which ones fail.

Artificial intelligence has now introduced a peculiar interruption to this process. Many of the tasks traditionally assigned to beginners are exactly the tasks AI performs well. Drafting, summarising, research assistance, basic coding, data analysis, administrative coordination, making presentations, documentation and taking notes in a meeting. Across industries, technology is proving remarkably effective at reducing the time required for routine cognitive work. From a productivity perspective, this is difficult to oppose.

If software can complete a task in ten minutes instead of three hours, organisations will understandably use it. The question is what disappears alongside the inefficiency. Entry-level work was never only work. It was an apprenticeship disguised as labour. A junior consultant building slides was not merely producing slides. She was learning how arguments are structured. A young programmer debugging code was not merely fixing errors. He was learning how systems behave under pressure. A researcher checking sources was learning what credibility feels like in practice.

The activity and the education were intertwined. Expertise was not purchased. It was produced. This is what makes the rise of artificial intelligence different from many previous workplace technologies. AI is not only automating work. It is increasingly automating the work through which people learned how to work.

Research, drafting, coding, documentation, analysis and presentation building have traditionally occupied the lower rungs of professional ladders. These tasks were often repetitive, but they exposed people to the realities of their profession. Through repetition they developed judgment. Through mistakes they developed understanding. Through observation they learned how experienced practitioners thought.

The immediate benefits are easy to see – productivity improves, costs fall, work moves faster, organisations gain the ability to produce more with fewer people. The longer-term consequences are harder to measure. Recent labour market data suggests that the first signs of this shift may already be appearing. Research from the United States Census Bureau has found declines in employment among younger workers in some occupations most exposed to generative AI. Employer surveys have similarly reported reduced demand for certain entry-level roles as organisations expand their use of AI tools. At the same time, businesses continue to demand workers who can exercise judgment, solve unfamiliar problems and make decisions under uncertainty.

This creates a challenge that receives far less attention than questions about automation and job losses. A labour market cannot endlessly consume expertise without also producing it. Every profession depends on a pipeline through which novices gradually become experts. Senior professionals do not emerge spontaneously. They are the result of years of accumulated experience. Remove enough of the experiences that traditionally created expertise and eventually the pipeline narrows.

The effects may not become visible immediately. Many organisations today are still drawing upon expertise developed under older conditions. The senior managers, partners, editors, engineers and consultants currently leading organisations learned their professions before generative AI became widely available. They acquired judgment through direct experience. They spent years performing the work that younger employees may increasingly bypass.

The real test may come a decade from now. Imagine a consulting firm in 2035. Its junior employees rarely build models because AI does it. They rarely conduct preliminary research because AI gathers information. They rarely draft reports because AI produces competent first versions. They spend more time reviewing, editing and coordinating than creating. The firm becomes more efficient. Projects move faster. Clients receive work more quickly. But a question gradually emerges. Who is becoming an expert?

In such a world, organisations could find themselves with large numbers of people who are highly skilled at managing AI systems but less experienced in the underlying work those systems perform. Everything may function smoothly when circumstances are predictable. The challenge will appear when something unusual happens.

This is why the future challenge may not be unemployment. It may be experience. The risk is not simply that AI replaces work. The risk is that societies become exceptionally good at generating outputs while gradually weakening the mechanisms through which expertise is formed.

None of this implies a rejection of technology. History offers little evidence that societies prosper by resisting useful tools. The printing press did not eliminate learning. Calculators did not eliminate mathematics. Search engines did not eliminate knowledge. Each technological advance changed what humans spent their time doing. Artificial intelligence will do the same. 

The challenge is ensuring that as machines assume more responsibility for routine cognitive work, societies do not accidentally dismantle the pathways through which people acquire judgment, experience and expertise.