I've been thinking about a possible long-term problem with AI that I don't hear discussed very much.
This isn't really about AI "taking our jobs." It's about how people become competent in the first place.
For most of history, learning and productive work happened at the same time.
A student wrote an essay because the teacher needed an essay to grade—but the real product was a student learning to organize and express an argument.
A new employee answered phones, worked in the mailroom, reconciled accounts, wrote simple code, prepared routine reports, or did other low-level work. The company needed the work done, but while doing it the novice learned how the organization and profession worked.
The immediate product wasn't the only product. The developing human was another product.
AI potentially breaks that relationship at both ends.
In education
A student who can't write very well used to produce a bad paper.
That failure was useful information. The student discovered that writing was difficult. The teacher could see where the student was struggling. And the only way for the student to get better was to struggle through more writing.
Now AI can produce a competent-looking paper for a student who cannot independently produce one.
So we potentially get a feedback loop:
Weak skills → greater AI reliance → less practice → weaker skills → still greater AI reliance.
AI didn't create our educational problems. Many of the trends were visible long before generative AI appeared. My concern is that AI may be an accelerant because it lets people bypass some of the cognitive work through which competence develops—and can conceal the fact that the competence is missing.
Then the same thing may happen in the workplace
Think about how people traditionally entered organizations.
Someone started in the warehouse, mailroom, answering phones, doing junior bookkeeping, basic research, routine drafting, simple programming, etc.
Those jobs weren't just inexpensive labor. They were the bottom rungs of a career ladder.
A capable 22-year-old did relatively simple work, learned from experienced people, made mistakes, gained judgment and gradually became the valuable 25-, 30- and 40-year-old professional.
AI is exceptionally good at a lot of that entry-level cognitive work.
From the perspective of an individual company, eliminating junior positions can make perfect economic sense. Why pay a new employee to spend three hours producing something an experienced employee can generate with AI in ten minutes?
But that creates a longer-term problem:
If AI eliminates the bottom rungs of the ladder, how does anybody get onto the ladder?
The company may not need the 22-year-old novice to do that routine work anymore.
But the company does need the 22-year-old novice to become the valuable 25-year-old team member it will need three years from now.
We've seen something similar before
Certain skilled trades provide an analogy.
When industries outsource work for decades, they don't merely outsource production. They can inadvertently outsource the process that creates skilled workers.
Experienced tool-and-die makers retire. There aren't enough younger journeymen behind them because there weren't enough apprentices doing the work ten or twenty years earlier.
You can buy new machinery and bring production back.
You can't instantly manufacture twenty years of human experience.
AI could potentially create the same problem across a much larger portion of the economy.
The traditional progression is:
Novice → routine work → mistakes → correction → harder work → experience → judgment → expert
If we automate the routine work, we risk creating:
Novice → AI → ??? → Expert
I'm not arguing that we should ban AI. Quite the opposite. AI may become one of the most powerful tools humans have ever developed.
I'm wondering whether we need to recognize that productive struggle has value independent of the immediate product being produced.
A seventh-grader may need to write an essay even though AI can write a better one.
A junior programmer may need to struggle through some code even though AI can generate it faster.
An apprentice may need to make something inefficiently while an expert watches.
Eventually we may deliberately assign humans work that machines can perform better—not because we need the work done that way, but because we need inexperienced humans to become experienced humans.
So here's the hypothesis I'd like people to tear apart:
Generative AI may accelerate pre-existing weaknesses in human-capital development by removing productive struggle at both ends of the pipeline: students can bypass cognitive work needed to develop foundational competence, while employers can automate entry-level work through which inexperienced adults historically converted foundational competence into professional expertise.
If that's true, the biggest long-term employment problem with AI might not be simply:
"What jobs will AI replace?"
It might be:
"If AI removes the bottom rungs of the ladder, where will the next generation of experts come from?"
School essays, junior office work and tool-and-die apprenticeships appear unrelated until you recognize that all three contain a hidden output: they manufacture experience.
I'm interested in arguments both for and against this. What am I missing?
