The specific thing happening in the job market is not mass layoffs. It is that the bottom rung of certain office careers is getting harder to climb onto, through hiring that slows rather than jobs that visibly disappear. If your plan was to get an entry-level job and work up, that plan carries more risk than it did for your parents. Here is what to build instead, starting from nothing.
What the data shows about entry-level hiring
Three separate research groups, using three completely different datasets, found the same shape. Payroll records covering millions of workers, resumes and job postings across 285,000 firms, and AI usage data all point to a decline in early-career employment in the occupations most exposed to AI.
Three datasets, same direction
Measured decline for early-career workers in AI-exposed occupations
The part that matters most for what you do about it: the mechanism is hiring, not firing. The Harvard study found separations actually went down slightly. Nobody is being pushed out. The door is closing behind the people already inside.
Two caveats worth carrying. Researchers genuinely disagree about how much of this is AI versus a post-pandemic hiring correction and high interest rates, and some credible groups find no difference between exposed and unexposed young workers at all. And it is early. Three years is not long enough to be certain about anything in labour markets. But three independent datasets pointing the same way is worth planning around even before the cause is settled.
Five moves to protect your income from AI
If you already have a job in an exposed field, do not quit it
Every study points the same way here. The damage falls on people trying to get in, not people already inside. Experienced workers in the same exposed occupations show no comparable gap over the period when young workers fell.
Leaving an exposed job to start over somewhere else means re-entering at exactly the rung that is disappearing. Experience is the asset. If you are going to move, move sideways with the experience intact, not down to the bottom of a new ladder.
This is the highest-value move on the list and it costs nothing.
Get paid while you train, rather than paying to train
Researchers at Microsoft scored occupations by how much of the work AI can currently do. Construction trades came in at 0.07 and home health and nursing assistants at 0.04, against 0.49 for interpreters and 0.45 for writers. The hands-on, in-person work sits at the bottom of every framework built so far.
The government projects the same shape: between 2024 and 2034, healthcare support grows 12.4% and construction 5.2%, against 3.1% for total employment, while office and administrative support loses 761,900 jobs.
Electricians specifically: high school diploma to enter, a paid four to five year apprenticeship, median pay of $62,350 in May 2024 (BLS has since published $63,190 for May 2025), and about 81,000 openings a year. For someone without capital, being paid to learn is structurally different from financing a degree and hoping.
Worth: a career with no debt attached, in the segment least exposed on every measure anyone has built.
Aim for work that pays for judgment rather than procedure
The Dallas Fed sorted occupations by how much more they pay once you have years behind you, then compared that against AI exposure. Occupations with no experience premium lost 0.28 percentage points of wage growth. Occupations in the top tenth gained 0.2 points.
The practical test: if the job could be learned entirely from a manual, it is exposed. If it takes years of judgment, or requires you physically present, it holds up better. That test cuts across the usual white-collar and blue-collar line.
Worth: a filter you can apply to any job posting in about ten seconds.
Use AI inside the job you have, rather than chasing an AI career
Research from the Washington Center for Equitable Growth found that augmentative uses of AI go with higher wages while automative uses go with lower ones. Same technology, opposite sign, depending on how it is used.
A caution on the numbers you will see: reports that job postings mentioning AI skills advertise 28% or 62% higher salaries come from analyses of advertisements, not from studies of what people actually get paid. Those postings skew senior, urban and toward high-paying firms. Treat AI fluency as a multiplier on skill you already have, not as a career in itself.
Worth: real, but smaller than the headlines suggest.
Build one source of income that does not depend on being hired
Here is a finding that gets overlooked. In a Danish study with administrative data on 25,000 workers, the researchers found no measurable effect on earnings or hours two years after ChatGPT launched, ruling out effects larger than 2%. So far, whatever time AI saves has not shown up in pay.[S1]
That is an argument for having something that pays you for what you own rather than what you do. It does not have to be property. A trade license you can bill under, a small business, a skill people hire directly, a rented-out room. The common feature is that the income does not require someone to decide to hire you.
For me that was a duplex. Renting out the other side meant my largest fixed cost stopped depending on my paycheck. That is not protection from AI and I am not going to dress it up as protection from AI. It is one bill that stopped being a bill, which changes how much a bad year can hurt you.
Worth: the whole point, and the slowest to build. Start it before you need it.
What I am not going to tell you
That property protects you from AI. No study I could find tests that, and anyone claiming it is making an argument rather than citing evidence, including me. What I can say is narrower: every finding above measures income from labour. Hiring rates, wage growth, exposure scores. None of it describes income that arrives because you own the thing producing it. Whether that distinction matters as much as it seems to is not something research has answered yet.
I am also not going to tell you the trades are easy, or that everyone should skip college. Recent graduates face 5.6% unemployment and 41.5% underemployment, and the unemployment advantage a degree carried in 1979 has inverted. That is a real change in the odds. It is not the same as the degree being worthless, and the research cannot make that call for your situation.
The one-paragraph version
The full evidence, including where researchers disagree, is here: what the data shows about AI and wages for young workers. If the part that worries you is your bills rather than your income, that is the other half: getting ahead of what AI is doing to housing costs.
Frequently asked questions
What does AI-proof income mean?
AI-proof income is defined here narrowly: income that does not depend on someone deciding to hire you, such as a trade license you can bill under, a small business, a skill people hire directly, or a rented-out room. No study tests whether property protects you from AI, so the claim is limited to income from what you own rather than income from labour.
Which jobs are least exposed to AI?
The least exposed jobs are hands-on, in-person work. Microsoft researchers scored construction trades at 0.07 and home health and nursing assistants at 0.04, against 0.49 for interpreters and 0.45 for writers. Government projections point the same way: healthcare support grows 12.4% and construction 5.2% between 2024 and 2034, against 3.1% for total employment.
Should you quit a job in an AI-exposed field?
No. Every study points to the damage falling on people trying to get in, not people already inside; experienced workers in exposed occupations show no comparable gap. Leaving means re-entering at exactly the rung that is disappearing, so move sideways with the experience intact rather than down to the bottom of a new ladder.
Is AI causing entry-level jobs to disappear?
The measured effect is slower hiring, not visible layoffs. Three research groups using payroll records, resumes and job postings across 285,000 firms, and AI usage data all found a decline in early-career employment in AI-exposed occupations, while separations went slightly down. Researchers still disagree on how much is AI versus a post-pandemic hiring correction and high interest rates.
Sources: Brynjolfsson, Chandar and Chen, Canaries in the Coal Mine, Stanford Digital Economy Lab; Hosseini Maasoum and Lichtinger, Harvard; Anthropic, Labor market impacts of AI; Tomlinson et al., Microsoft Research; US Bureau of Labor Statistics, Occupational Outlook Handbook and Employment Projections 2024-34; Federal Reserve Bank of Dallas; Federal Reserve Bank of New York; Humlum and Vestergaard, NBER Working Paper 33777; Washington Center for Equitable Growth. Figures current as of July 2026.
Keep going: if the first step is getting out on your own, start with moving out for the first time. If you are ready to look at owning, what house hacking actually is and the free calculators.
Rent from a tenant does not depend on your employer.
Free House Hacking Calculator. Put in a price and a rent and it shows how much of your housing payment a tenant would cover each month. That covered share keeps paying whether or not your job does.
Get started with the newsletter!
Useful, relevant analysis, tools, tricks, and what you need to know on the journey to financial freedom. From the van years to a duplex, the plain math and none of the hype.
Subscribe and get the free $0-to-First-Property Roadmap to start.
No spam. Unsubscribe in one click.
Want the deeper toolkit? The guides and spreadsheets are in the shop, from $5, and the free tools stay free either way.
Sources
- [S1] National Bureau of Economic Research (Anders Humlum and Emilie Vestergaard), ‘Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI’, Working Paper 33777 (revised March 2026), read 24 September 2026: “we estimate precise null effects on earnings and recorded hours at both the worker and workplace levels, ruling out effects larger than 2% two years after the launch of ChatGPT.” www.nber.org.