AI productivity gains are concrete and all signs point to this rising as the technology grows. But José Alberro of FTI Consulting argues that organizations can still benefit from AI gains and preserve the bottom rung of the employment ladder, allowing juniors to develop — and a talent crisis awaits if they don’t.
A particular kind of quiet occurs when a company stops replacing people instead of firing them. No press releases and no headlines come out. A departure gets absorbed instead of being backfilled. A graduate class shrinks from 40 to 12; nobody calls it a layoff, because it isn’t one. Multiply those scenarios across enough companies and enough years and you get something that never shows up as a spike in the unemployment rate, right up until it’s the defining fact of an entire generation’s early career.
That’s what AI appears to be doing to the labor market, not through mass layoffs but through a gradual narrowing of entry-level hiring that many firms — and most aggregate statistics — barely register. The instinct is to watch unemployment and layoff announcements for evidence of AI’s impact. but that effort is going to keep coming up empty, because that’s not where this impact shows up first.
Employment can fall for two very different reasons: Firms let people go, or they hire fewer people. An occupation can shrink for years through the hiring side alone — smaller graduate classes, unfilled vacancies or one person doing what used to take two — without a single person losing an existing job. Unemployment can stay calm. Wages can even look like they’re rising, because when the lowest rung of a job disappears first, whoever is left is, by definition, more senior and better paid, whether or not any individual received a raise. Aggregate numbers can tell a reassuring story while a real and specific group of people loses ground.
Same firms, same economy, opposite outcomes
Using matched employer-employee administrative data, a study by the US Census Bureau found that, following the introduction of ChatGPT, hires of workers 22 to 24 fell approximately 9% in the industries and states most exposed to AI, concluding that this reduction was the principal driver of the subsequent 12% decline in early-career employment. A recent Stanford Institute for Economic Policy Research (SIEPR) policy brief reached a more cautious but broadly consistent conclusion: Entry-level hiring in AI-exposed occupations has weakened, although AI is only one possible cause alongside higher interest rates, post-pandemic labor-market normalization and other structural changes.
Taken together, these studies suggest that AI’s first measurable labor-market effect has appeared less through mass layoffs than through a narrowing of entry-level hiring precisely where routine and codifiable tasks have traditionally provided the first rung of professional development.
The pattern is consistent across occupations and genders. Software development, a predominantly male occupation, and customer service, a predominantly female occupation, have both experienced weaker employment growth among younger workers. As reported by Fortune, evidence from the Stanford-ADP canaries dashboard suggests that although young women have experienced slower employment growth overall, most of the gap reflects occupational composition, as women are more heavily represented in occupations with greater AI exposure.
Why the first rung matters more than the ladder
Every knowledge profession runs on an apprenticeship that it rarely recognizes as one. Junior lawyers review documents so that five years later, they know what a bad contract looks like without being told. Junior consultants build the deck, so they eventually know what belongs in it. Junior coders write the boring functions so they can eventually architect the systems. Nobody assigns that work because it’s valuable. They assign it because that’s how someone becomes senior.
If AI increasingly performs that work instead, an organization becomes more efficient today but eventually stops manufacturing its own future experts. The irony is that this strategy may be individually rational yet collectively shortsighted. Every firm benefits from hiring experienced workers someone else trained, but if every firm reduces entry-level hiring at the same time, fewer experienced workers will exist a decade from now. The labor market may eventually discover it has optimized away the very pipeline on which professional expertise depends.
Payroll data don’t directly show this apprenticeship mechanism, but employment is weakening among precisely the workers whose jobs contain the largest share of routine, codifiable tasks. A 24-year-old who can’t get an entry-level version of a job isn’t just missing a paycheck this year, they’re missing the years during which judgment is built. Ten years from now, some firms may find they have plenty of AI and nobody left who grew up doing the work AI replaced.
None of this is an argument against adopting AI. Organizations that fail to deploy these technologies will almost certainly become less competitive over time. The productivity gains from AI are real and likely to increase as the technology matures. But organizations that capture AI’s efficiency gains while preserving effective pathways for developing junior talent are likely to enjoy both stronger short-term performance and a deeper pool of future expertise.
What’s worth watching
Rather than watching a long dashboard of labor-market statistics, two questions deserve sustained attention: Does new-graduate hiring keep shrinking relative to overall headcount at firms adopting AI tools? Does it take longer for younger workers to reach senior roles as the entry-level work through which expertise was traditionally acquired disappears?
AI systems will almost certainly become more capable, and firms will continue experimenting with ways to use less junior labor to produce the same output. The trend is already visible even if its ultimate magnitude remains uncertain.
What’s still to be determined is whether anyone rebuilds a way for a 23-year-old to become good at something when the work that used to get them there is gone. That’s a harder problem to solve than layoffs — and a much more subtle one.
AI productivity gains are concrete and all signs point to this rising as the technology grows. But José Alberro of FTI Consulting argues that organizations can still benefit from AI gains and preserve the bottom rung of the employment ladder, allowing juniors to develop — and a talent crisis awaits if they don’t.
A particular kind of quiet occurs when a company stops replacing people instead of firing them. No press releases and no headlines come out. A departure gets absorbed instead of being backfilled. A graduate class shrinks from 40 to 12; nobody calls it a layoff, because it isn’t one. Multiply those scenarios across enough companies and enough years and you get something that never shows up as a spike in the unemployment rate, right up until it’s the defining fact of an entire generation’s early career.
That’s what AI appears to be doing to the labor market, not through mass layoffs but through a gradual narrowing of entry-level hiring that many firms — and most aggregate statistics — barely register. The instinct is to watch unemployment and layoff announcements for evidence of AI’s impact. but that effort is going to keep coming up empty, because that’s not where this impact shows up first.
Employment can fall for two very different reasons: Firms let people go, or they hire fewer people. An occupation can shrink for years through the hiring side alone — smaller graduate classes, unfilled vacancies or one person doing what used to take two — without a single person losing an existing job. Unemployment can stay calm. Wages can even look like they’re rising, because when the lowest rung of a job disappears first, whoever is left is, by definition, more senior and better paid, whether or not any individual received a raise. Aggregate numbers can tell a reassuring story while a real and specific group of people loses ground.
Same firms, same economy, opposite outcomes
Using matched employer-employee administrative data, a study by the US Census Bureau found that, following the introduction of ChatGPT, hires of workers 22 to 24 fell approximately 9% in the industries and states most exposed to AI, concluding that this reduction was the principal driver of the subsequent 12% decline in early-career employment. A recent Stanford Institute for Economic Policy Research (SIEPR) policy brief reached a more cautious but broadly consistent conclusion: Entry-level hiring in AI-exposed occupations has weakened, although AI is only one possible cause alongside higher interest rates, post-pandemic labor-market normalization and other structural changes.
Taken together, these studies suggest that AI’s first measurable labor-market effect has appeared less through mass layoffs than through a narrowing of entry-level hiring precisely where routine and codifiable tasks have traditionally provided the first rung of professional development.
The pattern is consistent across occupations and genders. Software development, a predominantly male occupation, and customer service, a predominantly female occupation, have both experienced weaker employment growth among younger workers. As reported by Fortune, evidence from the Stanford-ADP canaries dashboard suggests that although young women have experienced slower employment growth overall, most of the gap reflects occupational composition, as women are more heavily represented in occupations with greater AI exposure.
Why the first rung matters more than the ladder
Every knowledge profession runs on an apprenticeship that it rarely recognizes as one. Junior lawyers review documents so that five years later, they know what a bad contract looks like without being told. Junior consultants build the deck, so they eventually know what belongs in it. Junior coders write the boring functions so they can eventually architect the systems. Nobody assigns that work because it’s valuable. They assign it because that’s how someone becomes senior.
If AI increasingly performs that work instead, an organization becomes more efficient today but eventually stops manufacturing its own future experts. The irony is that this strategy may be individually rational yet collectively shortsighted. Every firm benefits from hiring experienced workers someone else trained, but if every firm reduces entry-level hiring at the same time, fewer experienced workers will exist a decade from now. The labor market may eventually discover it has optimized away the very pipeline on which professional expertise depends.
Payroll data don’t directly show this apprenticeship mechanism, but employment is weakening among precisely the workers whose jobs contain the largest share of routine, codifiable tasks. A 24-year-old who can’t get an entry-level version of a job isn’t just missing a paycheck this year, they’re missing the years during which judgment is built. Ten years from now, some firms may find they have plenty of AI and nobody left who grew up doing the work AI replaced.
None of this is an argument against adopting AI. Organizations that fail to deploy these technologies will almost certainly become less competitive over time. The productivity gains from AI are real and likely to increase as the technology matures. But organizations that capture AI’s efficiency gains while preserving effective pathways for developing junior talent are likely to enjoy both stronger short-term performance and a deeper pool of future expertise.
What’s worth watching
Rather than watching a long dashboard of labor-market statistics, two questions deserve sustained attention: Does new-graduate hiring keep shrinking relative to overall headcount at firms adopting AI tools? Does it take longer for younger workers to reach senior roles as the entry-level work through which expertise was traditionally acquired disappears?
AI systems will almost certainly become more capable, and firms will continue experimenting with ways to use less junior labor to produce the same output. The trend is already visible even if its ultimate magnitude remains uncertain.
What’s still to be determined is whether anyone rebuilds a way for a 23-year-old to become good at something when the work that used to get them there is gone. That’s a harder problem to solve than layoffs — and a much more subtle one.
