The Apprenticeship Crisis Nobody Is Talking About

There is a question that keeps me up at night, and it has nothing to do with whether AI will take senior people's jobs. It is about the juniors.

Here is the pattern: agents are taking over the routine, repetitive, entry-level work in every knowledge profession. The first drafts. The research summaries. The data cleaning. The standard contract reviews. The boilerplate code. For organisations, this is a win — faster, cheaper, more consistent.

But that routine work was never just production. It was education.

How expertise was actually built

Nobody becomes an expert by reading about expertise. You become one by doing a thousand reps of the basic version, slowly developing the pattern recognition that lets you see what the textbook never taught you.

A junior lawyer who reviews two hundred contracts develops an instinct for which clauses matter and which are decoration. A junior analyst who builds fifty models starts to feel when the numbers are telling a story the formula cannot capture. A junior designer who produces three hundred variations learns taste — not from a framework, but from the accumulated experience of what worked and what did not.

That is apprenticeship. And it is being automated out of existence.

The ladder is being pulled up

When agents do the entry-level work, the entry-level job changes — or disappears. What remains requires the judgment and pattern recognition that only came from doing the work that no longer exists.

This is a structural paradox: the skills required for the roles that survive can only be built through the roles that are disappearing.

And I do not see enough people taking this seriously.

Organisations are celebrating the productivity gains without asking: where will the next generation of senior talent come from? Who is building the pipeline if the bottom rungs of the ladder are gone?

What universities are not doing

Most educational institutions are responding to this by adding AI courses. Learn prompt engineering. Understand large language models. Get certified in this tool or that platform.

This misses the point entirely.

The gap is not in tool proficiency. The gap is in the deep, domain-specific judgment that used to be built through years of practice — practice that AI is now doing instead. No course can substitute for the pattern recognition you develop by living inside a problem space for a decade.

What we need is not AI education. What we need is a fundamentally different model of professional development — one that accounts for the fact that the apprenticeship path is broken.

Three directions worth exploring

I do not have the full answer. But I see three directions that deserve serious attention.

Supervised agent collaboration. Instead of juniors doing the work, they supervise agents doing it — but with structured mentoring that surfaces the judgment calls the agent made invisibly. The learning happens in the review, not the production. This requires senior people to invest in teaching in a way most organisations do not currently value or reward.

Deliberate simulation. If the real work is too high-stakes or too scarce for practice, create structured environments where juniors face realistic complexity at compressed timescales. Medical training has done this with simulation labs. Law, consulting, finance — almost nowhere. AI could actually help build these simulations.

Explicit knowledge extraction. Senior professionals need to articulate the tacit knowledge they built through years of practice — the rules of thumb, the red flags, the "something feels off" instincts — and make them visible. Not as documentation that gets filed away, but as living frameworks that juniors can practice against.

The cost of doing nothing

If we do not solve this, the consequences compound quietly. Organisations will find, in five to seven years, that their senior talent is retiring and there is no one behind them with the depth to replace them. Not because young people are not smart or motivated, but because the path that built that depth no longer exists.

By then it will be too late to build it back.

This is not an AI problem. It is a human development problem. And it is the one that deserves our attention most.