Nobody’s Hiring the Bottom Rung
How AI Is Deleting the Entry-Level Job, and Breaking the Only Ladder a Career Was Ever Built On
The first jobs AI is taking aren’t the dangerous ones or the dull ones. They are the entry-levelones: the junior analyst, the first-year associate, the support rep, the junior developer. The work a person used to learn the trade by doing. Companies aren’t firing anyone for it, which is part of why it goes unnoticed. They’re doing something quieter and harder to see, simply declining to backfill the seat when it empties.
That sounds like a problem for new graduates, and it is. But it’s a deeper problem than that. Every senior professional alive climbed a ladder whose bottom rung was an entry-level job. Pull the rung out, and the people standing on it fall; so, eventually, does the ladder. In a decade, there’s no one left to promote.
Where the entry-level job is disappearing
Relative employment and hiring declines, by source
Published by Kymata Labs · Independent Research Institution.
You don’t have to be 22 to feel this one coming.
If you graduated in the last three years and the job market felt strangely closed, with applications vanishing into a void and “entry-level” postings that somehow demand three years of experience, this paper is about why. It wasn’t you. The bottom rung is being quietly sawed off.
If you’re a manager, ask a colder question: who’s your replacement? The junior you didn’t hire this year is the senior you can’t promote in 2034. The team looks lean and productive right now precisely because AI is doing the work a new hire used to do, and so, without quite deciding to, you are declining to grow anyone behind you.
And if you run a company, the trade in front of you is seductive and slow-acting: cut hiring costs today, defer a talent shortage to a CEO who isn’t you yet. The savings are immediate and on the balance sheet. The cost is deferred and off it.
A ladder rots from the bottom long before anyone standing near the top can feel it give.
“You can’t have senior people without junior people. We just stopped hiring the junior people.”
Kymata LabsForget the forecasts. This is already sitting in the payroll records.
The strongest claims here don’t rest on anyone’s prediction about the future. They come from what employers have already done, visible in payroll data, in leaked planning slides, and in the cold arithmetic of who is getting hired. Four findings, drawn from four independent sources, all point the same way.
The payroll data: the youngest workers, in the most exposed jobs, are vanishing first
The Stanford Digital Economy Lab went straight to the ledger, to the ADP payroll recordsrather than a survey, and found a pattern they titled, pointedly, “Canaries in the Coal Mine.” Since broad generative-AI adoption, early-career workers aged 22–25 in the most AI-exposed occupations have seen a 16% relative employment decline, even as older workers in the same jobs, and workers in less-exposed fields, held steady or grew. The adjustment runs through headcount, not wages: firms aren’t paying juniors less, they’re hiring fewer of them, and the effect concentrates exactly where AI automates the work rather than augments it. The sharpest single example: employment for software developers aged 22–25 has fallen roughly 20% from its late-2022 peak.
Brynjolfsson, Chandar & Chen, Stanford Digital Economy Lab, Nov 2025. Note: the Aug 2025 draft reported 13%; the figure was revised upto 16% in November. A Feb 2026 follow-up rebuts the “it’s just interest rates” counterargument.The leaked slide: a Big Four firm planning a third fewer juniors
This is the named corporate case, and it’s unusually concrete. An internal PwC US planning slide, reported by Business Insider, shows the firm intends to hire ~32% fewer entry-level associates by FY2028, dropping from 3,242 tax and assurance associates in FY2025 to 2,197 in FY2028, with audit cut 39%. The slide names the cause in the firm’s own language: “transformation efforts, the impact of AI, and further AC integration” (AC being the offshore Acceleration Centers). AI and offshoring, doing together what a thousand first-year associates used to do.
Business Insider, Aug 2025 (leaked internal slide + on-record PwC). The law-firm parallel is weaker: per NALP, AI has had “no impact on total law-grad hiring, yet.” So we lead with accounting, where the move is clear, and treat law as the likely next domino rather than one that has already fallen.The Fed’s ledger: new grads are now worse off than the workforce they’re joining
For most of modern memory, a fresh degree bought you a labor-market advantage. That has inverted. The New York Fed puts recent-graduate unemployment at ~5.7% in 2026:Q1, with 41.5% underemployed, working jobs that don’t require their degree. The St. Louis Fed sharpens the reversal: young grads aged 23–27 averaged 4.59% unemployment against a 3.25% comparable baseline. New entrants are now worse off than the broader workforce they are trying to join, and that is a historical flip, not a rounding error.
Federal Reserve Banks of New York (2026:Q1) and St. Louis (Aug 2025).Big Tech’s front door: new grads are now a rounding error of hiring
At the 15 largest technology companies, new graduates are now just 7% of hires. New-grad hiring is down roughly 25% from 2023 and more than 50% versus 2019. The industry that built itself on cheap, teachable, ambitious young engineers is, quietly, closing the door they came in through, at the precise moment its own tools made the junior engineer’s first-year tasks the easiest thing in the world to automate.
SignalFire, State of Talent Report 2025.The forecast, handled honestly: the warning that got loud, then quieter
In May 2025, Anthropic CEO Dario Amodei warned that AI could wipe out half of all entry-level white-collar jobs and push unemployment to 10–20% within one to five years. We include it because it framed the debate, and we flag it for exactly what it is: a forecast, not data. By May 2026, Amodei (and OpenAI’s Sam Altman) were reported to be softening those numbers. The honest position is the modest one. The prediction may yet prove high, but the things already measured (the payroll declines, the PwC slide, the Fed figures) don’t need the forecast to be alarming.
Amodei, via Axios, May 2025: a labeled forecast, later walked back, not a measurement.
New grads are now worse off than the workforce they’re joining
Unemployment rate, young grads vs a comparable baseline
The hard jobs were supposed to go first. The first jobs went instead.
The popular story had automation eating the difficult and the dangerous: the surgeon, the long-haul driver, the structural engineer. Reality ran the other way. Today’s AI is best at exactly the work we hand to beginners, the summarizing of a document, the drafting of a first pass, the reconciling of a spreadsheet, the boilerplate code, the routine ticket. The junior’s job description and the model’s sweet spot are nearly the same paragraph.
So companies make a rational, local decision. Why hire and train a first-year associate for a year of supervised reps when a model does the deliverable today, at a fraction of the cost, with no ramp? The Stanford data shows the answer playing out: the cut lands on headcount, not wages, and it lands hardest where AI automates rather than augments. The seat doesn’t get cheaper so much as it quietly stops existing.
Each of those decisions is defensible on its own. Stack them across an industry and you get something nobody chose: a profession that has stopped manufacturing its own future experts. The first-year associate was never just cheap labor. He was a senior partner sitting through an early, expensive, necessary draft of his own career.
We automated the apprenticeship and kept the title.
Two cohorts are forming, and the same tools hand them opposite fates.
One generation is already on the ladder: the mid-career and senior professionals who climbed when the bottom rung still existed. AI makes them more valuable, because they have the judgement to direct the model, the experience to catch it when it is confidently wrong, and the seniority to own the output. For them, the tool is pure leverage.
The other generation is arriving at the foot of a ladder that no longer has a first rung, asked to demonstrate experience the system has quietly stopped letting them earn. The 22-to-25-year-old in the Stanford data isn’t losing to a stronger candidate; the role itself was deleted before the application was ever filed. The line between the two cohorts is roughly the year you were born, and it is hardening into one of the more consequential divides of the decade.
The same data, read by three different readers.
A severed pipeline can be re-spliced, though not by wishing it. The same studies that diagnose the break also point at what holding it open would require. What that looks like depends on who you are.
For individuals
Aim past the rung that’s being deleted.
- Target roles where a human directs and checks the AI, the work that is surviving, rather than the routine output AI now produces outright.
- Get inside any apprenticeship that still exists, even a smaller one. Proximity to senior judgement is the asset; the title is not.
- Build the one thing AI still can’t be trusted with: judgement about when its confident answer is wrong.
For employers
You’re booking a saving today and a succession crisis later.
- An AI-lean team looks productive while quietly training no one, the same trap PwC’s own slide describes, at scale.
- Treat junior hiring as R&D on your own talent, not a cost line. The first-year associate is next decade’s partner, or there isn’t one.
- Redesign the entry role around supervising AI, so the rung still exists, just higher up the ladder than it used to be.
For policymakers
The labor market is reversing on the people you most need to protect.
- New grads are now worse off than the broader workforce, a historical flip the Fed data already shows.
- Fund apprenticeships and earn-while-you-learn paths directly; the private pipeline that used to provide them is closing.
- Watch accounting and consulting as the leading edge, and law as the likely next domino, and measure it before the ladder is fully gone.
FAQ
It's a fair challenge, and the honest answer is that the two are tangled together. Junior roles are cyclical; they soften when budgets tighten. But the Stanford Digital Economy Lab isolated the AI signal: the 16% decline shows up specifically for 22-to-25-year-olds in the most AI-exposed occupations, while older workers in the very same jobs, and workers in less-exposed fields, held steady or grew. A pure downturn doesn't sort that cleanly by age and exposure. And the lab's own Feb 2026 follow-up directly rebuts the "it's just interest rates" reading. The cycle is real, but it isn't the whole story.
One firm, but a loud one, and the slide is specific: ~32% fewer entry-level associates by FY2028, with audit cut 39%, attributed in PwC's own words to "transformation efforts, the impact of AI, and further AC integration." The macro data is what turns one slide into a pattern: SignalFire finds new-grad hiring down more than 50% versus 2019 at the largest tech firms, and the Stanford figures span occupations, not companies. We've been deliberate about where the evidence is strong. Accounting and consulting are clearly moving. Law has not moved yet, and we say so.
Some did, and we say so plainly. Dario Amodei's May 2025 warning (half of entry-level white-collar jobs gone, unemployment at 10–20% within five years) is a forecast, not a measurement, and by May 2026 he and Sam Altman were reported to be softening it. So we don't lean on the forecast. We lean on what's already in the payroll records: a 16% relative employment decline that has happened, not one that might. Think of the forecast as the ceiling of the worry and the measured data as its floor, and notice that even the floor is high enough to be alarming.
Whoever needs a senior in ten years. Seniority is not a credential you buy; it's junior work, survived and metabolized. The first-year associate who reconciles the messy ledger becomes the manager who knows when a number is lying. Delete the rung and you don't just lose this year's hires; you stop manufacturing next decade's experts. The bill arrives late, which is exactly why it's easy to ignore now.
Get inside an apprenticeship pipeline that still exists, even a smaller one, and become the person who supervises the AI rather than the person it replaces. The roles disappearing are the ones AI automates outright, while the roles that survive are the ones where a human directs, checks, and owns the model's output. Aim for that second kind early, and treat judgement, the thing AI still can't be trusted with, as the skill you're actually being hired to build.
Someone has to do the junior work. Right now, nobody’s being hired to learn it.
This isn’t an argument against using AI for the routine work. It’s genuinely good at it, and pretending otherwise helps no one. It’s an argument for noticing what the routine work was for. It was the classroom. Automate the lesson and keep the diploma, and you save a fortune now, then wake up one day with no one qualified to run the place. The cheapest seat in the building was always the most important one.
Hire the bottom rung. It’s the only part of the ladder that builds the rest.
Sources
Every figure in this paper is drawn from the primary sources below. Where a number was revised, where the evidence is a forecast rather than a measurement, or where a parallel is weaker than the headline case, we have said so in the text.
- Brynjolfsson, E., Chandar, B. & Chen, R. (2025). "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence." Stanford Digital Economy Lab (Nov 13, 2025). Data: ADP payroll records. (An Aug 2025 draft reported a 13% decline; revised to 16% in Nov 2025. See also the Feb 2026 follow-up, "Canaries, Interest Rates, and Timing.") https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- Business Insider (Aug 21, 2025). "PwC plans to hire far fewer junior associates as AI and offshoring reshape the Big Four" (leaked internal slide plus on-record PwC). https://www.businessinsider.com/pwc-hiring-fewer-junior-associates-ai-offshoring-big-four-2025-8
- Federal Reserve Bank of New York. "The Labor Market for Recent College Graduates" (2026:Q1: recent-grad unemployment ~5.7%, underemployment 41.5%). https://www.newyorkfed.org/research/college-labor-market
- Federal Reserve Bank of St. Louis (Aug 2025). "Recent College Grads Bear the Brunt of Labor Market Shifts" (young grads 23–27 averaged 4.59% unemployment vs a 3.25% comparable baseline). https://www.stlouisfed.org/on-the-economy/2025/aug/recent-college-grads-bear-brunt-labor-market-shifts
- SignalFire (2025). "State of Talent Report 2025" (new grads now ~7% of Big Tech hires; new-grad hiring down ~25% vs 2023 and >50% vs 2019 at the 15 largest tech companies). https://www.signalfire.com/blog/signalfire-state-of-talent-report-2025
- Axios (May 28, 2025). "Anthropic CEO Dario Amodei: AI could wipe out half of all entry-level white-collar jobs." (A labeled forecast. By May 2026, Amodei and OpenAI's Sam Altman were reported to have softened these predictions.) https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic
Read it as a PDF
The complete argument, the cross-domain evidence, and all primary sources, typeset as a white paper. Free, no signup.
↓ Download the PDF