AI-Adopting Companies Are Still Hiring. Entry-Level Jobs in AI Fields Are Vanishing.
Workers aged 22 to 25 in AI-exposed occupations are running 19% behind their peers on employment, according to Revelio Labs, while the companies that actually adopted AI keep adding headcount.
Key Takeways
- Workers aged 22 to 25 in AI-exposed occupations are 19% behind their peers on employment, the widest shortfall Revelio Labs measured for any group.
- Firms that adopted AI are recording fewer layoffs and expanding headcount relative to firms that have not, which puts the damage in hiring rather than firing.
- The information sector lost 23,000 jobs in August 2026, roughly three times its 8,000 average monthly loss over the trailing 12 months, per the Bureau of Labor Statistics.
- A company that stops hiring 23-year-olds this year will have no 33-year-olds with a decade of AI-era judgment in 2036.
A 23-year-old sending out applications for entry-level jobs in an AI-exposed field this month is running about 19% behind peers her age who applied somewhere else. Same graduation year, a different door. That figure comes from Revelio Labs, published September 3, and it is the widest employment shortfall the firm measured across any group it tracks.
The Bureau of Labor Statistics published its August employment report the next day, and the headline read calm. Nonfarm payrolls rose 162,000 against a consensus forecast of 53,000. Unemployment held at 4.1%. June and July were both revised upward, by 11,000 and 44,000. On the aggregate measures, August was a solid month, and BLS did the careful work that lets anyone say so with confidence.
The information sector is shedding jobs at three times its own recent pace
Underneath that total, one sector moved differently. Information lost 23,000 jobs in August, roughly three times the 8,000 it has averaged in monthly losses over the trailing 12 months. Information is also where AI adoption runs highest: 39.7% of firms in the sector, against a national average of 19.8%, per Census Bureau data. Without the BLS sector series, that detail disappears into the total and nobody sees it at all.
| U.S. Bureau of Labor Statistics, August 2026 | Jobs |
|---|---|
| Information sector jobs lost in August | 23,000 |
| Information sector average monthly loss, trailing 12 months | 8,000 |
| Total nonfarm payrolls added in August | +162,000 |
One month proves little on its own, which is why the running version of these numbers lives on the AI Job Loss Statistics hub rather than in any single headline. Last month's story ran the other direction: AI-attributed layoffs hit a five-month high while total layoffs hit a two-year low.
The firms using AI the most are cutting the fewest people
Revelio Labs counts jobs its own way, through its Revelio Public Labor Statistics and its AI Labor Market Tracker, and its August count was 36.5K. That comes from a different method than the BLS payroll survey, so the two describe the month in their own terms. The shape underneath Revelio's numbers is where it gets interesting. Employment in the most AI-exposed occupations sits about 6% below the least-exposed ones, cumulative since November 2022. AI-exposed firms are recording fewer layoffs than the least-exposed firms. And firms that have actually adopted AI keep expanding headcount relative to the ones that have not.
The labor market added very few jobs in August, and opportunities are becoming increasingly uneven. That is especially visible for younger workers in AI-exposed occupations.
Lisa Simon is Revelio's chief economist, and her firm's own reading of its evidence puts the weight somewhere narrower than a broad wave of AI-driven job destruction: on who gets hired into AI-exposed roles, especially the youngest applicants, and on which companies are growing. That is a careful thing for a data company to say in a week when the louder version would have traveled further.
The on-ramp is where the damage is concentrated
Set the two releases beside each other and a specific picture appears. Companies deep into AI are growing. Companies that have not adopted are cutting harder. And the people absorbing the cost are the ones who have not started yet: 22 to 25 years old, first real job, applying into the exact fields where AI is being deployed fastest. If your junior hiring looks thinner this year than last, you are in good company, and that is the part worth sitting with.
Every experienced person I have worked with earned their judgment the same way. They were wrong early, on something small, with someone senior close enough to catch it. 20 years inside a Fortune 100 taught me that the senior people a company will need in 2036 are sitting in its junior seats right now, learning by being corrected. Close that door for 3 years and the correction never happens.
The people who can tell when the model is wrong take a decade to grow
AI capability keeps converging toward a common floor, and the tools every competitor can rent look more alike each year. Subtract the identical AI from both sides of a market and what remains is whatever each company actually owns: process knowledge nobody wrote down, and the people who can tell when the output is wrong. I made that argument at length in a piece on the layoffs that buy the one thing every competitor will have too. A workforce is the hardest advantage to copy, because it is the one that walks out the door the moment you cut it.
Call that layer Above the Algorithm: judgment, taste, trust, accountability. None of it arrives fully formed. It accrues, slowly, through supervised mistakes. A company that stops hiring 23-year-olds this year has decided it will have no 33-year-olds with a decade of AI-era judgment in 2036. Closing an on-ramp takes a quarter. Rebuilding that kind of judgment takes 10 years, and the two decisions get logged on the same headcount line as though they moved at the same speed.
Zoom out from the hiring plan and this is a conversation happening at a few hundred thousand kitchen tables. Someone moves home, takes something adjacent, and starts the real career 2 years late. Careers compound from the first rung, so a delay there bends the whole curve, and it bends wages and the tax base along with it. The companies keeping that door open while everyone else closes it are making a talent bet their competitors are not.
So here is the question worth carrying into your next leadership meeting. How many people under 26 did we hire into AI-exposed work this year, and what did we spend teaching them to judge what the model hands back? If nobody in the meeting knows the number, you have your answer. And if you are the 23-year-old on the other side of this, the question belongs to you too: ask the company interviewing you who trained the last junior they hired, and listen to how quickly the answer comes. I am easy to find if you want to compare notes.
What counts as an AI-exposed occupation?
Exposure measures how much of a job's actual tasks overlap with what AI systems can do now. Whether the employer has adopted AI is a separate question. That separation is why Revelio Labs can report weak employment in AI-exposed occupations and strong hiring at AI-adopting firms in one release. A paralegal at a firm with no AI tools counts as exposed. A warehouse planner at a company running AI across operations may not.
Why do two jobs reports for the same month show different numbers?
The Bureau of Labor Statistics surveys employers and households on a set schedule and publishes payroll counts from that work. Revelio Labs builds its own count from its Revelio Public Labor Statistics and AI Labor Market Tracker. The methods differ and the underlying data differs, so neither one corrects the other. Comparing the two headline counts side by side will mislead you.
Should new graduates avoid AI-exposed fields?
The August 2026 data points the other way. AI-adopting employers are the ones still expanding headcount, so the more useful screen is the employer rather than the field. Ask whether the company hired anyone junior in the last year, and ask who supervised that person's work. The answer tells you whether there is a seat where you can be wrong early and learn from it.
Here is what makes Alex a credible voice on this topic: Alex advises the California State University system on AI governance and shaped Cisco's $1.1B innovation portfolio across 20 years, two vantage points on the question this post raises, which is who trains the people who will judge AI's output in 2036.
For more, book a conversation →
.svg)
