The employment statistics look reassuring. Unemployment in most developed economies has remained near historic lows even as AI adoption has accelerated. The catastrophic mass displacement that some predicted — offices emptied, professions collapsed, white-collar workers queueing at unemployment offices — has not materialized. If you're measuring the AI transition by job elimination, the numbers are not alarming.
But that's the wrong measurement. The question is not how many jobs are disappearing. The question is what is happening inside the jobs that remain.
The Anatomy of a Hollowed Job
Consider a financial analyst at a mid-size asset management firm in 2019 versus 2025. In 2019, the analyst's work involved gathering data from multiple sources, building Excel models, synthesizing research, writing memos, and presenting recommendations. These tasks required judgment, domain expertise, and a significant investment of cognitive effort.
In 2025, the same analyst has the same title, the same office, and a salary that has grown with inflation. But the data gathering is now automated. The initial model is AI-generated. The first draft of the research synthesis is produced by an LLM. The analyst's job has become reviewing, correcting, and approving AI output — a fundamentally different cognitive task from the work the role was designed around.
This is hollowing. The job exists. The compensation is roughly maintained. But the work has changed in ways that have significant implications for skill development, professional identity, and long-term career trajectory.
What Gets Lost When the Work Changes
The tasks that AI has absorbed first — research, synthesis, first-draft creation, data processing — are not just administrative burdens. They are also the mechanisms through which junior professionals develop expertise.
A junior lawyer who spends three years doing document review is not just billing hours. They are building pattern recognition — learning to notice the detail that changes a case, developing the intuition about what matters that experienced lawyers call judgment. When AI does the document review, the junior lawyer skips the work and also skips the formation process.
Economists call this "task displacement" as distinct from "job displacement." The job persists; the tasks within it change; the developmental pathway that the tasks enabled disappears. A 2024 study of law firms using AI document review found that associates reported feeling less certain of their own judgment two years into AI-assisted work than they had at the end of law school. The work that built confidence and capability had been optimized away.
The same pattern appears in medicine, accounting, engineering, and any field where AI has taken over well-defined analytical tasks. The remaining work is either higher-stakes judgment that requires the expertise AI can't match, or supervisory oversight of AI systems — neither of which develops the foundational expertise in the same way the automated tasks did.
The Oversight Problem
There's an additional dynamic that doesn't get enough attention: the cognitive demands of AI oversight are genuinely different from the cognitive demands of doing the underlying work, and not necessarily simpler.
Supervising an AI system that produces plausible-sounding outputs requires a specific kind of critical awareness — the ability to catch errors that are subtle, context-dependent, and not flagged by any internal error signal. This is, in some respects, harder than doing the work directly, because it requires both knowing the answer and knowing what confident-sounding wrong answers look like.
Most professionals have not been trained for this. Medical training teaches diagnosis; it does not specifically train doctors to detect the failure modes of AI diagnostic systems. Legal training teaches analysis; it doesn't specifically train lawyers to catch the ways LLMs confidently misstate case law. The oversight role requires expertise that was built through doing — which is precisely the work that's been automated away.
The result is a peculiar form of professional deskilling that is invisible in the aggregate statistics. Workers are nominally employed in roles that require their expertise. But the mechanism for developing and maintaining that expertise has been disrupted.
The Compensation Question
For now, hollowed jobs often pay comparably to the more substantive versions that preceded them. The labor market hasn't yet fully priced the difference between a lawyer who does deep legal work and a lawyer who reviews AI output. That gap will close.
As AI systems improve, the oversight role becomes more tractable for less experienced (and less expensive) workers. The premium that experienced professionals command rests partly on their accumulated expertise — and that expertise becomes less differentiating when AI does the work that used to require it. The hollowing of jobs is, on a long enough timeline, also the hollowing of the wage premium that comes with expertise.
A 2025 NBER working paper found early evidence of this compression: in fields with high AI adoption, wage growth for mid-career professionals had slowed significantly relative to comparable fields with lower adoption, while entry-level wages remained stable. The expertise premium was deflating.
The Identity Dimension
There is a dimension of job hollowing that doesn't appear in economic data at all: what happens to people when the work that gave their career meaning is the work that gets removed.
A radiologist who spent fifteen years developing the ability to read complex imaging — who derived genuine satisfaction from the moments of expert pattern recognition that made a diagnosis — is in a different psychological position than the economist's model of a worker whose productivity has been augmented. The augmentation has taken something. Not the job. Not the compensation. The specific form of engagement that made the work feel like more than labor.
This is not a trivial concern. Identity, meaning, and the relationship between work and self-worth are not externalities to a functioning labor market. They are the lived experience of the people inside it. The hollowing of jobs is also, in ways that are harder to measure, the hollowing of professional lives.
After Work takes these losses seriously — not as sentimental objections to progress, but as real costs that should be counted, understood, and addressed in the design of the systems that are replacing the work.