The Apprenticeship Trap
AI is eliminating the grunt work that built every senior professional you know
Here is a partial list of people who started their careers doing work that AI can now do faster and cheaper: the CEO of Goldman Sachs, who started as an analyst building pitch books. The managing partner of every major law firm, who spent years reviewing contracts by hand. The chief technology officer of most Fortune 500 companies, who learned to code by debugging other people’s mistakes. The senior partner at McKinsey, who built slide decks at 2 a.m.
In every case, the grunt work wasn’t incidental to their expertise. It was the mechanism that created it. The tedious, repetitive, low-status labor that fills the early years of a professional career is not just a hazing ritual. It is the training ground. And AI is closing it.
The Numbers
Entry-level job postings in the United States have fallen roughly 35 percent since January 2023, according to labor research firm Revelio Labs. At the 15 largest tech companies, hiring of new graduates has dropped more than 50 percent since 2019, per venture capital firm SignalFire. The decline is consistent across functions: sales, marketing, engineering, recruiting, operations, design, finance, and legal all saw the same contraction.
Two-thirds of global enterprises now plan to cut entry-level hiring due to AI adoption and automation. PwC UK has explicitly cited AI in cutting roughly 200 entry-level roles. A Citi and Hildebrandt report found that 86 percent of large law firms plan to grow their associate ranks through 2027, but only 35 percent plan to increase the size of their first-year classes. The gap tells the story: firms want more senior lawyers, not more junior ones.
For the first time since tracking began in 1990, the unemployment rate for recent college graduates now exceeds the national average. Computer science graduates face 6.1 percent unemployment. Computer engineering graduates, 7.5 percent. Philosophy majors? 3.2 percent. The degrees that were supposed to be safest are getting hit hardest, precisely because they trained people for the tasks AI handles best.
The Apprenticeship Model
To understand why this matters beyond the immediate job market pain, you have to understand how expertise actually develops in professional industries.
Finance, law, consulting, medicine, accounting, software engineering, and architecture all share a common structure. Junior professionals enter and spend years doing unglamorous work under the supervision of senior people. The work itself is the curriculum. A first-year analyst builds financial models not because the firm can’t automate it, but because building models is how you learn to understand and use them. A junior associate reviews thousands of contracts not because partners enjoy delegating drudgery, but because pattern recognition only develops through repetition.
This is the apprenticeship model, and it has governed professional development for centuries. The senior people at the top of every professional hierarchy got there by going through it. They learned judgment by first learning mechanics. They developed intuition by first developing fluency with the raw material.
As one MIT researcher put it: the way you make a senior employee is not through school. It’s by doing the job alongside someone who knows more, and you learn by doing. That’s where the bulk of our skill comes from.
AI doesn’t just automate junior tasks. It removes the training mechanism that produces senior professionals. That distinction is everything.
Industry by Industry
Finance. Wall Street firms are reportedly considering pulling back junior analyst hiring by as much as two-thirds. The tasks being automated are exactly the ones that constituted the apprenticeship: building pitch books, running valuation comparisons, updating financial models, screening compliance data. The irony is acute. Junior analysts at Goldman Sachs famously work 80 to 100 hours per week, and much of that work is now AI-addressable. But those hours were the forge. The banker who can understand a deal at a glance learned by building hundreds of models from scratch.
Law. Nearly half of attorneys at large firms now use AI tools. Document review, legal research, due diligence, basic drafting: these are the bread and butter of junior associate work, and they are increasingly automated. One junior associate told MIT Technology Review that he worries he’s not getting the same repetitions that senior attorneys got. Law firm leaders acknowledge the concern. When junior work dries up, as one Wharton professor noted, you have to find a more formal way of teaching than hoping that an apprenticeship works. But nobody has built that replacement yet.
Technology. U.S. programmer employment fell 27.5 percent between 2023 and 2025, according to the Bureau of Labor Statistics. The roles vanishing fastest are the entry points: debugging, testing, routine maintenance, basic feature implementation. These were not merely jobs. They were how software engineers developed the understanding of systems that eventually qualified them to architect those systems. You cannot design a complex codebase if you have never spent years inside one fixing things that broke.
Consulting. The slide deck that a junior consultant builds at 2 a.m. is not just a deliverable. It is the process by which that consultant learns to structure an argument, synthesize data, and present it to a client who will challenge every assumption. AI can generate a deck. It cannot generate the judgment that comes from having built and defended hundreds of them. Some firms are increasing hiring, others are cutting. The split likely reflects which firms have thought through the pipeline problem and which are simply optimizing for short-term cost.
Accounting. Audit, tax preparation, bookkeeping, and compliance work are among the most automatable professional tasks. These are also the entry points into a profession that leads to CFO roles, controller positions, and advisory practices. The Big Four have been early AI adopters. The question nobody is asking publicly: where do the next generation of partners come from if the first five years of the career path no longer exist?
The Barbell
What emerges across all these industries is the same structural shape: a barbell. On one side, senior professionals who built their expertise the old way, through years of apprenticeship-style work. On the other side, AI tools that can perform many of the tasks those professionals learned on. The middle disappears.
In the short term, this looks like a productivity gain. Firms can do more with fewer junior employees. Margins improve. Headcount drops. Earnings calls celebrate efficiency.
In the medium term, it becomes a crisis. The senior professionals who currently carry institutional knowledge, client relationships, and decision-making authority will retire. The pipeline that was supposed to produce their replacements has been narrowed or closed. The firms that cut junior hiring in 2024 and 2025 will find themselves, by 2030 and 2035, with a talent gap they cannot fill by hiring laterally, because every competitor made the same cuts.
This is not a hypothetical. It is a predictable, structural shortage being created in real time by rational short-term decisions that ignore medium-term consequences. That is, by definition, a second-order effect.
The On-Ramp Problem
Zoom out from any single industry and the picture gets worse. Finance, law, consulting, and technology have historically served as the professional on-ramps for ambitious college graduates. These were the industries that absorbed the top of the talent distribution, trained them intensively for a few years, and then distributed them across the broader economy. A former investment banking analyst might become a startup CFO. A former associate at a law firm might become general counsel at a mid-size company. A former consultant might run operations at a healthcare system.
All four on-ramps are narrowing simultaneously. Entry-level postings are declining across all of them. The traditional path from elite university to professional services to broader leadership is being disrupted at the entry point, with no clear alternative emerging.
The optimists say these graduates will simply skip the grunt work and start at a higher level, augmented by AI. This misunderstands what the grunt work actually was. You cannot augment judgment that was never developed. You cannot shortcut pattern recognition that requires thousands of hours of exposure. AI can give a first-year analyst the output of a financial model. It cannot give them the understanding of why the assumptions matter, which only comes from having built models with wrong assumptions and watching them fail.
The historical analogies people reach for are misleading. When spreadsheets replaced manual accounting, the work of accounting didn’t disappear. It shifted to higher-level analysis, but the entry path remained. You still learned by doing. When legal databases replaced physical law libraries, junior associates still did research, just faster. The apprenticeship survived because the tools augmented the work without eliminating the need for a human to do it.
AI is different because it doesn’t just speed up the junior professional’s work. In many cases, it replaces the need for a junior professional to do the work at all. The senior partner can prompt the AI directly. The training step gets skipped entirely.
Who Pays the Cost
The cost of this transition will not be distributed evenly.
Students from wealthy families with professional networks can find alternative paths into senior roles through connections, unpaid internships, and family-funded gap years spent building portfolios. Students from working-class and middle-class backgrounds relied on the traditional pipeline: get the degree, get the entry-level job, learn on the job, advance on merit. If the entry-level job disappears, the meritocratic ladder disappears with it.
The university system, which has spent decades selling the promise that a degree is the ticket to a professional career, faces an existential problem it has not yet reckoned with. The value proposition of a four-year degree was always built on the assumption that entry-level professional employment would be there on the other side. If it isn’t, the $200,000 investment in a bachelor’s degree becomes much harder to justify, particularly for students taking on debt to finance it.
And the broader economy faces a question nobody seems to be asking: what happens to consumer spending, household formation, and social stability when an entire generation of college graduates enter the workforce with no clear path to the professional careers they were promised? The political implications alone are significant. A generation locked out of the traditional economic ladder will not remain quietly frustrated.
Nobody’s Job to Solve
The reason this problem isn’t being addressed is structural. No single institution has an incentive to solve it.
Companies are optimizing for quarterly earnings and cutting junior headcount while deploying AI improves margins immediately. The talent pipeline problem is a 2030 issue, and most executives are evaluated on this quarter’s performance. Universities are still selling degrees based on historical employment data that no longer reflects the market their graduates will enter. Reforming curricula is slow, politically fraught, and threatens the revenue model. Government has barely begun to register the problem, let alone develop policy responses.
The institutions that could intervene (professional associations, licensing bodies, industry groups) are largely captured by the interests of current practitioners, who benefit from reduced competition at the junior level. Fewer junior lawyers means less billing competition for senior partners. Fewer junior analysts means more leverage for experienced professionals. The people with the power to rebuild the pipeline have a personal financial interest in not doing so.
This is a coordination failure of the most classic kind. Every actor is behaving rationally from their own perspective. The collective result is a system that is consuming the seed corn of its own future talent supply.
The System Is Eating Its Seed Corn
The standard AI narrative has two tracks. The optimistic version: AI will create more jobs than it destroys, and the workforce will adapt. The pessimistic version: AI will destroy millions of jobs and we need universal basic income. Both tracks miss the specific problem outlined here.
The apprenticeship pipeline is not about job quantity. It is about knowledge transfer. It is about how a society reproduces its professional expertise from one generation to the next. Even if AI creates net new jobs, and it may, the question remains: where do the next generation of senior professionals come from if the mechanism for training them no longer exists?
The firms celebrating their AI-driven productivity gains today are drawing down an account they did not build and cannot replenish on the current trajectory. The senior talent they rely on was produced by a system they are now dismantling. The savings are real. The cost is deferred. And deferred costs, in every system from finance to agriculture to infrastructure, eventually come due.
We are not just automating tasks. We are automating the process by which people learn to do the tasks that cannot be automated or to understand the output of automated tasks. If that sentence seems circular, that’s because the problem is. And circular problems do not resolve themselves.
Second Order Effects
Policy analysis through the lens of unintended consequences and incentives.