AI Labor and Formation

The Erosion of Apprenticeship

AI does not only replace tasks. It can erase the awkward, inefficient work through which novices become competent. A field that automates its junior layer may look productive for a decade while quietly destroying its own future experts.

The most important labor question in the AI transition is not simply, “How many jobs disappear?” It is, “Which jobs disappear first?”

In many knowledge fields, the first work to be automated is the work that used to train beginners: drafting, summarizing, checking, tagging, searching, formatting, cleaning data, writing first versions, answering routine questions, preparing briefs, reviewing logs, debugging simple problems, making mistakes under supervision, and watching how seniors reason.

That work is often boring. It is also formative.

The junior task is not only an output unit. It is a training environment. When firms automate the output, they may also automate away the environment where skill used to grow.

The Hidden Function of Entry-Level Work

Entry-level work has always served two functions at once.

The visible function is production. The junior analyst builds the spreadsheet. The junior lawyer searches cases. The resident writes notes. The new programmer fixes small bugs. The editorial assistant prepares copy. The research assistant cleans data.

The hidden function is apprenticeship.

The beginner learns what good work looks like by touching low-risk pieces of real work. They learn the vocabulary, the pace, the shortcuts, the standards, the politics, and the judgment of the field. They learn when a senior is worried. They learn what errors look like before those errors become expensive. They learn how to recover.

That learning is difficult to formalize because much of it is tacit. It does not live in the handbook. It lives in the repetition of supervised work.

AI attacks the visible function first. The hidden function is collateral damage.

The Evidence So Far

The evidence is still developing, but the pattern is visible enough to treat seriously.

Microsoft’s 2026 Future of Work synthesis points to uneven benefits from AI and highlights a longer-term concern: automating jobs that help workers learn skills may undermine how expertise is built. It cites evidence that employment for workers aged 22 to 25 in highly AI-exposed jobs declined relative to similar less-exposed work, and that junior hiring slows after firms adopt AI.

Stanford-linked analysis of payroll data reported that young workers in AI-exposed occupations, including software development and customer support, have seen sharper employment losses than older workers in comparable fields. Several press summaries put the decline for young workers in highly exposed roles around the low-to-mid teens since late 2022, while older workers in some exposed categories held up better or grew.

The World Economic Forum’s 2025 Future of Jobs Report expects major labor-market churn by 2030, with employers planning both AI-driven restructuring and significant reskilling. That sounds adaptive, but reskilling cannot fully substitute for apprenticeship if the worksite no longer contains real beginner tasks.

At the same time, the evidence is not one-sided. Studies of generative AI in customer support and other settings have found that AI can improve productivity most for novices or lower-skilled workers by making tacit knowledge more available. AI can be a tutor, scaffold, and equalizer.

That is the tension. AI can help beginners learn. It can also give managers an excuse not to hire them.

The Seniority Trap

Organizations under pressure tend to make a rational short-term choice:

This can look brilliant on a quarterly dashboard. Output rises. Headcount falls. Senior employees become more productive. Clients still receive deliverables. Investors hear a credible efficiency story.

Then the lag appears.

Five years later, there are fewer mid-level people who learned the work properly. Ten years later, the senior people are retiring or burning out. The field discovers that it saved money by consuming the seed corn.

This is the seniority trap: AI lets an institution overuse the existing expert layer while underproducing the next one.

Verification Without Formation

One proposed future is that humans will become reviewers, editors, validators, and supervisors of AI output. That is plausible, but it hides a formation problem.

You cannot become a good reviewer merely by reviewing. You become a good reviewer by first doing the work badly, then less badly, then competently, while someone better shows you what you missed.

A junior programmer who never writes simple code may struggle to review AI-generated code. A junior lawyer who never drafts rough arguments may struggle to see why a polished AI memo is misleading. A junior journalist who never reports may accept smooth summaries without noticing missing ground truth. A junior doctor who overrelies on generated notes may lose the habit of close observation.

Verification is not a lower-skill substitute for production. In many fields, verification is a higher-order skill built on production experience.

If AI removes the practice path, it also weakens the review path.

Fields At Risk

The apprenticeship problem is most acute where three things are true:

Software is an obvious case. So are law, finance, consulting, accounting, journalism, design, marketing, academic research, customer support management, data analysis, and parts of medicine.

The danger is not identical in every field. Medicine has licensing and supervision structures. Law has professional liability. Software has faster feedback but weaker gatekeeping. Journalism has public trust collapse. Academia has citation and peer-review problems. Finance has model risk and compliance.

But the shared pattern is the same: the tasks easiest to automate are often the tasks beginners need.

What Good Institutions Should Do

The answer is not to preserve every obsolete task. Some work should disappear. No one needs to defend meaningless drudgery for its own sake.

The answer is to replace accidental apprenticeship with deliberate apprenticeship.

A serious AI-era institution should maintain:

The best use of AI is not to remove the novice from the work. It is to let the novice see more examples, get faster feedback, test more drafts, and understand expert reasoning sooner.

That requires design. It will not happen automatically.

The Spiralist Reading

Apprenticeship is a human continuity system. It is how a field remembers itself through people rather than documents.

AI changes the archive of work. It can store examples, compress instruction, explain patterns, and simulate practice. But a field is not only an archive. It is a living chain of attention. Seniors notice things. Juniors absorb standards. Peers compare mistakes. Bodies sit in rooms. People learn what matters by seeing what others refuse to ignore.

When that chain breaks, the loss may not show up immediately. The dashboards will still glow. The outputs will still arrive. The summaries will be clean. The models will be helpful.

But fewer people will have been formed.

That is the quiet institutional danger. A society can have more output and less competence. It can have more tools and fewer craftspeople. It can have more answers and fewer people who know how an answer becomes trustworthy.

The Spiralist position is simple: do not let efficiency consume formation.

Practical Rule

For any AI-heavy field, ask:

What work did beginners used to do here?
What did that work teach them?
If AI now performs it, where does that learning happen instead?
Who is responsible for making sure it happens?

If an institution cannot answer those questions, it is not modernizing its apprenticeship system. It is liquidating it.

Sources and Context