AI Is Closing the Door on Entry-Level Jobs

The story everyone expected was mass layoffs. A company announces AI, thousands of people lose jobs, the headline writes itself.

That is not what the data shows. The actual pattern is quieter and harder to protest: the jobs are not being cut, they are never being posted.

A Stanford analysis of payroll records found a 13% relative employment decline for workers aged 22 to 25 in the most AI-exposed occupations, while older workers doing the same jobs held steady or grew. Nobody was fired. The entry rung was simply removed.

Here is what the evidence supports, what it does not, and what it means whether you are hiring or job hunting.

Key Takeaways

• Stanford payroll research found a 13% relative employment decline for 22 to 25 year olds in the most AI-exposed roles, with older workers unaffected.
• Recent graduates aged 22 to 27 faced 5.6% unemployment against an overall rate of 4.2%.
• 23% of first-quarter layoffs cited AI automation in regulatory filings, up from 14% the previous quarter.
• Employers explicitly attributed 54,836 job losses to AI in a year, while models estimate the real figure at 200,000 to 300,000.
• Research suggests many cuts are made on AI's expected capability rather than its demonstrated performance.

What the Data Actually Shows

The Stanford finding is the most useful because it compares like with like.

Researchers examined payroll records and separated workers by age within the same occupations. In roles most exposed to AI, employment for 22 to 25 year olds fell 13% in relative terms since late 2022. Workers in older brackets, doing those same jobs, held steady or increased.

That design matters. If AI were simply replacing a function, everyone doing it would be affected. Only the youngest cohort moving, in the same roles, points at something specific: the tasks being automated are disproportionately the ones juniors were hired to do.

Supporting figures line up:

• Recent graduates aged 22 to 27 had 5.6% unemployment against 4.2% overall
• 23% of first-quarter layoffs explicitly cited AI automation in regulatory filings, up from 14% the prior quarter
• About one in six employers expect AI to reduce headcount this year
• Wall Street banks plan to remove roughly 200,000 roles over three to five years, concentrated in entry-level and back office work

The scale is genuinely contested. Employers explicitly attributed 54,836 job losses to AI in a year, but modelling that includes positions never created puts the figure at 200,000 to 300,000. That gap between stated and estimated is the whole problem: a job that is never posted does not appear in any layoff statistic.

Why Juniors Specifically

Entry-level work in professional services has a common shape. It is well-defined, heavily reviewed, and volume-based: summarise documents, reconcile records, draft first versions, pull data into a template, check outputs against a rule.

That description is almost exactly the capability profile of current AI systems. A model operating software can complete structured, bounded, high-volume work faster than a person, which is the change we covered in what actually changed with the shift to computer use.

Senior work has a different shape. It involves judgment under ambiguity, client relationships, accountability for outcomes, and deciding which question to ask. Those resist automation for now.

So the automation lands unevenly by design, and the youngest cohort absorbs it. The junior role was the automatable part of the profession.

There is a structural problem hiding in that sentence. Senior people were produced by junior work. Remove the training rung and the pipeline that generates experienced staff goes with it, on a delay long enough that nobody feels it during the quarter they made the decision.

The Part That Complicates the Story

Before treating this as settled, one finding deserves weight.

Research published in Harvard Business Review argues companies are laying off workers because of AI's potential rather than its demonstrated performance. The cuts are being made against what executives expect the technology to do, not against measured productivity gains already achieved.

That reframes things considerably. If a firm cuts twenty junior roles expecting AI to cover the work, and it does not, the firm has a capability gap it created. The layoff still happened, and it still shows up as AI-attributed, but the underlying justification was a forecast.

This matters for anyone reading the trend. Some portion of these cuts is real substitution and some is anticipation, and the data does not cleanly separate them. Companies that cut too early on a forecast will be hiring again, quietly, in a couple of years.

It also explains why the explicit and estimated figures diverge so widely. Attribution is partly a narrative decision. "We are restructuring around AI" is a better message to investors than "demand softened".

What to Do About It

If you are early in your career

Target the judgment, not the task. Work that involves deciding what should be done is holding up far better than work that involves doing a defined thing well. Volunteer for the ambiguous problem rather than the clean one.

Become the person who verifies output. When a model does 200 records in minutes, someone has to design the check that catches the error. That skill is scarce, newly valuable, and most teams have not built it.

Look where AI exposure is lower. The 13% decline was concentrated in the most exposed occupations. Roles involving physical presence, regulated sign-off, or direct relationship management show a different pattern.

If you are hiring

Work out whether you are cutting on evidence or expectation. If you cannot point to a measured productivity gain, you are forecasting. That may be right, but name it accurately.

Price the pipeline. Removing junior roles saves salary now and creates a senior shortage in three to five years, when hiring experienced staff externally costs considerably more than developing them did.

The governance side of deploying these systems is covered in what to ask before AI agents ship into your stack, and the adoption pressure in why businesses are racing to deploy them.

Harvard Business Review published the analysis on layoffs driven by AI's potential rather than its performance.

Frequently Asked Questions

Is AI actually taking jobs right now?

The evidence supports a narrower claim than mass replacement. Stanford payroll research found a 13% relative employment decline for workers aged 22 to 25 in the most AI-exposed occupations, while older workers in those same roles held steady. The effect concentrates at the entry level rather than across the workforce.

How many jobs have been lost to AI?

It depends on how you count. Employers explicitly attributed 54,836 job losses to AI in a year, but modelling that includes roles never created estimates 200,000 to 300,000. The gap exists because a job that is never posted appears in no layoff statistic.

Why are entry-level jobs most affected by AI?

Entry-level professional work tends to be well-defined, heavily reviewed, and volume-based, which closely matches what current AI systems do well. Senior work involves judgment under ambiguity, client relationships, and accountability, which resists automation. The junior role was the automatable part.

Are companies cutting jobs because AI works or because they expect it to?

Research in Harvard Business Review argues many cuts are based on AI's expected capability rather than demonstrated performance. That means some portion of AI-attributed layoffs reflects a forecast rather than a measured productivity gain, and firms that cut too early may quietly rehire.

What jobs are safest from AI right now?

Roles centred on judgment under ambiguity, physical presence, regulated sign-off, or direct relationship management show far less impact than structured, high-volume work. Verifying and quality-checking AI output is also an emerging area where demand is rising rather than falling.

The Bottom Line

The thing to watch is not the layoff announcement. It is the job posting that never appears, because that leaves no trace in any statistic and no one to interview about it.

A 13% relative decline for one age group, in the same roles where older workers held steady, is the clearest evidence available that the entry rung is being removed rather than the profession being replaced.

Whether you are starting out or making the hiring decision, the useful question is the same: which part of this work is judgment, and which part is volume? The answer is deciding who gets hired.