Something has changed in the market for first jobs, and the data now shows it plainly. Employment among workers aged 22 to 25 in the most AI-exposed occupations sits about 19% below trend. The benchmark for that trend is employment among similarly aged workers in occupations AI barely touches. That figure comes from the Stanford Digital Economy Lab, which tracks payroll records from ADP, the provider that runs payroll for millions of US workers. The reading covers data through June 2026 and is measured against a November 2022 baseline. A year earlier the same gap was 15%.
The researchers are careful about what this does not mean, and so should anyone quoting the number.
It does not mean 19% of entry-level jobs have disappeared. It is the distance between two trajectories. In absolute terms, employment for 22-to-25-year-olds in the two most AI-exposed quintiles fell roughly 11% between November 2022 and June 2026, while employment for the same age group in the three least-exposed quintiles grew roughly 10%. Add those movements together and you get the gap. One group shrank, another grew, and the space between them is what the headline figure describes.
Two further qualifications matter. The adjustment runs almost entirely through reduced hiring rather than through layoffs, so this is a story about doors opening more slowly, not about people being pushed out. And the researchers describe their findings as descriptive patterns rather than causal estimates: they report no widespread, economy-wide job displacement, and they stress that payroll data alone cannot separate generative AI from interest rates, hiring freezes and ordinary cyclical caution.
So the honest summary is narrower and more useful than the headlines. The bottom rung has become harder to reach in specific occupations, while the ladder itself is being rebuilt in a different shape. This guide covers what the 2026 evidence actually shows, which entry paths remain open, and how to build proof of ability when no employer will hand you a first job.
Key Takeaways
- The AI employment gap for workers aged 22 to 25 widened from 15% to 19% between July 2025 and June 2026. It measures the distance between two trajectories, not a 19% loss of jobs.
- Recent graduate unemployment stayed high at 5.6% in the second quarter of 2026, with 42% underemployed.
- Employers still project a 5.6% increase in Class of 2026 hiring, so the market is weak rather than closed.
- More employers report losing junior learning tasks to AI than report losing junior drudgery.
- Critical thinking and communication outrank AI literacy in what employers say they want from a first hire.
What the 2026 Data Shows About AI and Entry-Level Jobs
Three independent sources describe the same market from different angles. Read together, they explain why the experience of job hunting feels worse than the aggregate numbers suggest.
The Employment Gap for the Youngest Workers
The Stanford Digital Economy Lab tracks employment by age band and by how exposed an occupation is to generative AI. Its finding is a divergence, not a collapse. Young workers in highly exposed occupations have lost ground relative to young workers elsewhere, and the mechanism is hiring: firms are opening fewer junior positions rather than removing the people already in them. The decline concentrates in roles built on codified knowledge, meaning formal, documented information that a model can reproduce.
The revised paper draws out that distinction more sharply than the original. Employment fell among young workers in occupations that lean on codified knowledge, the kind that can be taught through education, textbooks or written procedures. It rose among experienced workers in occupations that lean on tacit knowledge, the kind acquired through practice, mentorship and repetition. AI absorbed what could be written down. What resisted is what has to be lived through, which is precisely the thing a first job used to provide.
Experienced workers in the same occupations show a different pattern. Where AI complements the work rather than substituting for it, employment has stayed flat or risen. The split matters for your search: the risk sits in specific task profiles, not in whole industries.
A Weak Graduate Market That Is Not Closing
The Federal Reserve Bank of New York puts unemployment for recent college graduates at 5.6% for the second quarter of 2026, with an underemployment rate of 42%. Underemployment here means holding a job that does not normally require a degree, such as a graduate working a retail shift. Both figures point to a market where finding work is hard and finding work that uses your degree is harder.
The employer side reads slightly better. The National Association of Colleges and Employers surveyed 185 organisations for its Job Outlook 2026 Spring Update, published in April 2026. Those employers expect to hire 5.6% more graduates from the Class of 2026 than from the class before it. Modest growth against high underemployment is an uncomfortable combination, but it is not a closed door.
Where AI Removes the First Rung, and Where It Does Not
The useful question is not whether AI takes entry-level jobs. It is which parts of an entry-level job it takes first.
Survey work by Strada Education Foundation gives a concrete answer. It collected responses from nearly 1,500 executives and senior talent leaders and was published in May 2026. Among respondents, 42% said AI had increased the analytical and judgment-based responsibilities of entry-level staff. A third, 33%, said it had reduced their routine and administrative tasks. And 41% said it had reduced the foundational skill-building tasks that junior employees traditionally learn on.
Compare those last two figures. More employers report losing the learning tasks than report losing the drudgery, which is the opposite of the story usually told about automation. Photocopying the deal file, reconciling the ledger and drafting the first version of a memo were never valuable in themselves. They were how people learned the shape of the work. Remove them and the job gets more interesting and much harder to start. Employers who think seriously about AI augmentation rather than blanket automation tend to notice this earlier than those chasing headcount savings.
The pattern also has a distributional edge. The losses land on the people who have no track record to fall back on. That is one of the mechanisms discussed in work on automation and inequality, and it is why this is not only a career question.
What Employers Say They Want From a First Hire
The same Strada survey asked what senior talent leaders expect from AI over the coming year. In total, 46% expect AI use to increase entry-level hiring in 2026, against 17% who expect it to cut hiring. That is a ratio of 2.7 to one in favour of growth. Among firms that reported growing entry-level headcount, 27% named greater use of AI in the organisation as the most significant factor.
The skills ranking is more surprising. Asked to rate what matters in an entry-level candidate, respondents put critical thinking and communication highest, at 4.3 on the survey’s five-point importance scale. AI literacy came last, at 3.6. Employers also ranked candidates with direct work experience in a similar role as most desirable, and candidates with a 4.0 grade point average but no work experience as least desirable.
Read that carefully before you spend six months collecting certificates. Tool fluency is table stakes, not a differentiator. What employers say they cannot find is someone who can frame a problem, judge whether an answer is plausible and explain the reasoning to a colleague. If you have used AI inside a real business process and can describe what it got wrong and how you caught it, that story is worth more than the certificate.
The New Shape of the First Job
Where the traditional junior role has thinned out, something closer to an apprenticeship has started to replace it. The common features are visible across employers that are still hiring at the entry level: a longer supervised period, explicit review of AI-assisted output, and a narrower initial remit that widens as judgment develops.
This changes what a good first job looks like. Instead of asking how quickly you can be given real responsibility, ask how much structured feedback the role includes. A position where a senior colleague reviews your work weekly will teach you more in a year than one where you are left to prompt a model unsupervised. If the team has no obvious mentor, ask whether the company runs a mentorship matching programme.
It also changes how careers move afterwards. Progress increasingly looks like a career lattice rather than a straight climb, with lateral moves into adjacent functions doing the work that promotions used to do. Companies that run internal talent marketplaces or systematic cross-training make those moves easier to plan, so it is worth asking about both in an interview.
AI Skills Now Pay Outside the Technology Industry
One finding cuts against the assumption that you need an engineering path. The labour-market data firm Lightcast released an analysis in July 2025 called Beyond the Buzz: Developing the AI Skills Employers Actually Need. It found that job postings requiring AI skills advertised salaries about 28% higher than comparable postings without them, a difference of nearly $18,000 a year. It also found that 51% of those postings sat outside IT and computer science occupations.
Marketing, finance, operations, human resources and customer support are all buying AI capability now. That widens the set of realistic entry points considerably. It also means the fastest route into AI-adjacent work often runs through a function you already understand rather than through a computer science degree.
The practical implication is to pair a domain with a tool. Someone who understands accounts receivable and can automate part of the reconciliation is more employable than someone who knows the tool alone. The same logic explains why newer remote-first roles often combine a business function with technical fluency rather than separating them.
How to Build Proof When No One Hands You the First Rung
If employers prefer direct experience to grades, the task is to manufacture defensible experience before anyone employs you.
Ship Small, Finished Work
One completed project beats five abandoned ones. Pick a problem small enough to finish in a fortnight, build it, and write down what you tried, what broke and how you verified the result. Add version control and a short README, the file that explains what a project does and how to run it. An honest note on limitations signals the judgment employers say they are missing.
Where AI is part of the work, show your review process rather than the output. Describe the prompt, the wrong answer it produced first and the check that caught it. That is the skill the Strada respondents rated at 4.3.
Use the Side Doors
Several established routes get you a track record without a permanent offer. Remote internships compress real project work into a few weeks and now run at organisations that would not otherwise take juniors. Contract-to-hire arrangements let an employer test you on a live problem, which is precisely the evidence they say they want. Short, assessed credentials also matter more than they used to as skills-first hiring and micro-credentials spread. Micro-credentials are short courses that end in an assessment rather than a certificate of attendance, and they work best attached to a portfolio rather than instead of one.
Do not overlook people who have already left. Corporate alumni networks are one of the few referral routes open to candidates with no internal contacts, and referrals still move applications faster than anything you can write.
A Ninety-Day Plan for an AI-Exposed Job Market
A search without structure turns into scrolling. Give yourself three phases and a deadline for each.
Weeks one to four: narrow the target. Choose one function and one industry. Read twenty real postings and list the tools, tasks and phrases that repeat. That list, not a generic curriculum, is your study plan. Check advertised ranges as you go, since pay transparency rules now put salary bands in many postings and give you a baseline before any conversation about money.
Weeks five to eight: build the evidence. Complete one project that uses those exact tools on a problem from that exact industry. Document it. Then rewrite your CV so each bullet names an action, a tool and a result, in that order, and delete anything that does not.
Weeks nine to twelve: apply narrowly and follow up. Twenty tailored applications beat two hundred generic ones. Most large employers filter with AI hiring tools before a person reads anything, so mirror the wording of the posting instead of inventing synonyms for it. Reference a specific responsibility from the posting in your covering note, record the date you applied, and send one short follow-up after ten working days.
- Track applications, replies and interviews so you can see which step is failing.
- Ask in every interview how junior work is reviewed and by whom.
- Treat a rejection with feedback as more valuable than silence from a bigger employer.
- Confirm that an older posting is still live before investing time in it.
What Employers Can Do to Rebuild the Ladder
The hiring side of this problem is solvable, and the organisations that solve it will have a mid-level bench in five years when their competitors do not.
Start by naming which foundational tasks AI has absorbed. Then replace the learning they used to provide with deliberate structure: rotations, paired review, and written explanations of decisions rather than only their outputs. Talent pipeline partnerships with education and training providers make the intake more predictable than reactive posting.
Internally, keep movement possible. Clear internal mobility policies and flexible arrangements such as job sharing widen the set of people who can take a first role in a new function. Broader questions about how the gains from automation are distributed, including the long-running debate over universal basic income, sit outside any single company’s control. Building a workplace that adapts to automation does not.
Conclusion
The evidence for 2026 supports neither panic nor complacency. A measurable gap has opened for the youngest workers in AI-exposed occupations, graduate underemployment is high, and the tasks juniors used to learn on are thinning out. At the same time, employers expect to hire more graduates than last year. Most talent leaders expect AI to raise entry-level hiring rather than cut it, and AI-related pay premiums are showing up well outside the technology sector.
What that combination rewards is specificity. Pick a function, learn the tools that function actually uses, finish something small enough to explain, and apply to twenty employers who need exactly that. The first rung is harder to reach than it was three years ago. It has not been removed.
Found this useful?
Make SmartKeys a preferred source on Google, and our articles will surface more often in your Top Stories, AI Overviews, and AI Mode.
Add as Preferred Source
