Executive Summary

The belief that AI is cutting entry-level hiring is everywhere in the labor market. When Indeed Flex asked roughly 2,000 workers in the US and UK, 62% said so. Yet only 31% said they had personally seen that change in their own field. Belief runs exactly twice as far as experience.

That does not mean the squeeze is imaginary. When Stanford analyzed data from the largest US payroll processor, employment among 22-to-25-year-old early-career workers in the most AI-exposed jobs had fallen 13 to 16%. A survey measures what people feel; a payroll ledger measures how many were actually hired. When the two gauges disagree, the one that moves policy and hiring first is usually the louder, more alarming feeling.

What this moment really calls for is reading emotion data and measured data side by side. Whether the same AI shrinks junior hiring in one place or grows it in another comes down to a single axis: automation versus augmentation.

62%

Workers who believe it is shrinking

Indeed Flex survey (2,000 US·UK)

31%

Workers who actually experienced it

Half the level of belief

13-16%

Drop in young hiring in AI-exposed jobs

Stanford payroll data (ages 22-25)

65%

Firms hiring the same or more juniors this year

Hiring plans stay steady

1

62% and 31%: Two Numbers in One Survey

In July 2026, the staffing platform Indeed Flex asked roughly 2,000 workers in the US and UK how AI is changing hiring (PR Newswire). 62% said "AI is replacing routine tasks, so companies are cutting entry-level hiring." Yet only 31% said they had personally experienced that change in their own field. The number who believe is twice the number who have seen it.

Other numbers in the same survey are just as telling. Asked about the biggest obstacles in the job market, respondents still named pay (30%) and competition (25%) first, with AI ranking below both. 34% said AI had made them less confident about competing in today's labor market. What these figures capture is not the actual hiring ledger but the temperature of people's anxiety. What the survey measures is feeling, and fear usually runs ahead of the statistics.

Put the survey that measures feeling next to the payroll ledger that measures actual hiring, and the gap becomes visible. The left side of the figure below is the survey perception we just saw; the right side is the payroll ledger we examine later. First see how differently the two describe the same phenomenon of "AI and entry-level hiring."

Same phenomenon, different gauges Emotion data (survey) 62% believe it is shrinking 31% have experienced it Indeed Flex survey ~2,000 US·UK workers Measured data (payroll) −13-16% ages 22-25 hiring stable senior workers Stanford · ADP payroll data high AI-exposure jobs
▲ A survey measures how people feel; a payroll ledger measures how many were hired. The two gauges describe the same phenomenon differently — original Pebblous diagram
2

Inside Companies, the Story Sounds Different

Employer-side surveys have their own texture. HR Dive summarized a ResumeTemplates survey of 1,000 hiring managers and a joint Cognizant-Pearson study: 48% said they would rather invest in AI tools than hire and train new college graduates, and 55% had already shifted part of their entry-level hiring budget to AI tools.

More striking is the optimism the same managers hold. In this research, 94% of HR leaders said AI will create new entry-level roles within five years, and 96% expected the junior role to shift toward managing and supervising AI. The hand moving budget to AI now and the mouth promising to reopen junior roles later coexist inside a single organization.

College graduates in caps and gowns gathered at a commencement ceremony — the young jobseekers at the center of the debate over whether AI is cutting entry-level hiring
▲ A commencement ceremony packed with new graduates. Employer budgets are shifting toward AI, but survey data shows hiring plans themselves have not shrunk nearly as much | Source: Wikimedia Commons (CC BY 2.0)

And in the same research, 65% said they plan to hire the same number of new graduates as last year, or more, this year. Budget is shifting to AI, but the hiring door has not closed. What's more, companies' biggest complaint about new hires was not AI but the fundamentals. Work ethic, basic document comprehension, and professional email writing came up more often than AI. Even the employer voice sits at a distance from the simple story that "AI is eliminating entry-level jobs."

3

But the Payroll Data Really Did Move

Up to this point everything is survey data, that is, about what people feel and plan. What does the actual hiring ledger say? Brynjolfsson and colleagues at the Stanford Digital Economy Lab analyzed individual-level data from ADP, the largest US payroll processor. The paper is titled "Canaries in the Coal Mine."

Since generative AI spread in earnest in late 2022, employment of 22-to-25-year-old early-career workers in the most AI-exposed jobs fell by a relative 13 to 16%. Among software developers alone, the drop reaches about 20%. By contrast, employment of senior workers in the same jobs, and of workers in jobs with low AI exposure, stayed stable. The adjustment came through hiring less, not through cutting wages. Fear had simply run ahead, but it was not without grounds.

Stanford University's Gates Computer Science Building — home to the Stanford Digital Economy Lab, whose payroll-data study underlies this section
▲ Stanford's Gates Computer Science Building. The "Canaries in the Coal Mine" study cited above comes from the Stanford Digital Economy Lab's analysis of ADP payroll data | Source: Wikimedia Commons (Public Domain)

That said, even this measured data leaves causation in dispute. Some researchers argue the drop in young employment may not be AI's doing but an artifact of a cyclical hiring slowdown overlapping with age-cohort effects. Being measured does not make the interpretation automatically true. That is precisely why the numbers need to be read apart.

4

Why the Same AI Yields Opposite Outcomes

So why does the same AI shrink junior hiring at one company and grow it at another? The fork lies in whether AI replaces human work or assists it. Used to replace (automation), it lowers the number of juniors needed for execution work; used to assist (augmentation), juniors become productive faster, so hiring holds or rises.

Same AI, two employment outcomes Same AI tool Automation (replace) AI does the execution work Junior hiring ↓ Stanford · résumé data Augmentation (assist) AI supports junior judgment Hiring held · ↑ ECB survey · IBM case What you use AI for splits the direction of junior hiring
▲ The same AI shrinks junior hiring when it replaces execution work, and holds or grows it when it assists junior judgment — original Pebblous diagram

This fork is not only theoretical. A study of 65 million résumés found that firms adopting generative AI hired fewer juniors than non-adopters, with the decline concentrated in jobs with high AI exposure. Conversely, when the European Central Bank surveyed 5,000 firms, those using AI intensively for growth and R&D actually raised hiring by 4 percentage points more. Firms using it to cut costs hired less; firms using it to expand capability hired more.

IBM is the case that shows this axis head-on. In early 2026 IBM announced it would triple entry-level hiring, redesigning junior roles around customer engagement and complex problem-solving while shrinking the share of AI-replaceable work like coding. Faced with the same AI, one company treats it as a reason to shed juniors, another as a reason to hire more.

5

How to Read Emotion Data and Measured Data Apart

The data on this topic splits into at least three layers: the anxiety workers feel (survey), the plans employers state (survey), and how many were actually hired (payroll and job-posting ledgers). Lump the three together and you get the one-liner that "AI is destroying jobs"; pull them apart and different stories emerge. Here is a short checklist for taking in the news on the job.

  • 1. Label it first. Decide whether the number is what people feel (survey) or what actually happened (payroll, postings). 62% is a feeling; 13 to 16% is a measurement.
  • 2. If it is measured, ask whether it is automation or augmentation. The same drop or rise calls for a different response depending on whether AI replaced people or assisted them.
  • 3. When the two tracks disagree, that is itself a signal. If feeling points to a squeeze while measurement stays calm, it may be overblown fear; the reverse may be a risk not yet felt.

Blur the point where perception and measurement diverge, and both policy and hiring get dragged toward the louder, emotional side. Placing the two datasets side by side and checking which gauge measures what is the most practical line of defense for reading labor-market news in the age of AI.

R

References

Primary Reporting

Academic Research

Industry & Market Sources