Young college graduates in the United States are having a harder time getting work than they did a few years ago. If you manage a remote team, that is not only a headline about AI taking entry-level jobs. It is a reminder that juniors do not pick up the job by sitting near people anymore, unless you set that learning up on purpose.
The junior market has softened, and AI is not the only story.
Among U.S. workers age 22 to 26 who hold a bachelor's degree, the employment-population ratio was 88% in September 2026, a full percentage point below where it was three years earlier and nearly a point and a half below its immediate pre-pandemic level, according to Ernie Tedeschi at Stripe Economics (Nov 12, 2026). That group's unemployment rate was close to 6% at the end of 2025, versus around 4% before the pandemic, though Tedeschi notes it appears to be cooling. Unemployed young graduates are also searching longer: the typical spell was 12 weeks, a third higher than a few years ago.
It is fair to ask whether AI is shrinking the first rung of the ladder. Tedeschi says the evidence on AI's employment effects is still early and inconclusive, and that the hypothesis is worth taking seriously. He also shows why blaming AI entirely does not fit. Citing Ozimek and Goldschlag (2026), he notes that youth employment is also down since 2023 for young people without a college degree, who are less likely to be in AI-using industries or AI-exposed jobs. Those non-graduate rates have fallen slightly more since December 2019 than young graduate rates, while mid-career workers age 30 to 49 have returned to pre-pandemic levels.
Stripe sees a work-from-home pattern. The authors call it suggestive, not causal.
Tedeschi's own analysis of Current Population Survey data compares 2023, when recent-graduate employment rates peaked, with the 12 months ending in September 2026. In occupations with more work from home, employment for young graduates grew more slowly than for mid-career workers. For every extra 10 percentage points of work-from-home adoption, employment rates for young college graduates grew between 3.9 and 6.6 percentage points more slowly than those for middle-aged workers. The relationship was statistically significant.
As an accounting exercise, not a causal one, Stripe asked how much of the 1.23-percentage-point decline in the young-graduate full-time-equivalent employment-population ratio between 2023 and 2026 work from home could explain. The answer was 0.69 to 1.08 percentage points, or 56% to 88% of the decline. A measure of occupational AI exposure showed a weaker relationship that was statistically indistinguishable from zero. When Stripe included both work from home and AI exposure in the same model, the work-from-home effect stayed significant and the AI measure did not.
Those are Stripe's findings, and the authors are careful about what they are not. "Our model is suggestive rather than truly causal," Tedeschi writes. AI exposure and the ability to work remotely are highly correlated, which makes the two hard to separate. Other shocks were happening at the same time, and the full effect of AI, he adds, might not show up until the next recession. This article follows that caution. It is not a brief for a return-to-office mandate. Tedeschi also reviews evidence that remote work has helped disabled workers and older workers, including Bloom, Dahl, and Rooth (2026) in the United States and Burn, Gentry, and Lahey (2026) in the United Kingdom. Those gains are real. The question for an owner is how to keep a path for juniors without throwing those gains away.
Juniors learn from feedback. Remote teams have to arrange it.
The mechanism Stripe suggests is straightforward: without the informal learning that happens in person, employers are likelier to hire older workers for jobs where working from home is now expected. Tedeschi points to Emanuel, Harrington, and Pallais (2026), who find that physical proximity to the office increases the job feedback software engineers receive, especially younger and less-experienced workers.
You do not need a new study to recognize the pattern. A junior who can lean over a cubicle wall gets corrected in the moment. A junior in another city gets a message, a shared document, or silence. Some of the quick questions that used to create that contact now go to an AI tool instead of a colleague, a shift we described in Your team stopped asking each other questions. That's a problem.. The cost is how a new person learns what "good" looks like in your business.
AI can also hide the problem by making first drafts look finished. If a junior can produce a passable memo or a spreadsheet with a model, a busy manager may skip the review that used to be the training. In AI gave your team back a third of the time on the tasks it took, the point was that time saved by AI does not automatically become better judgment. Someone still has to decide where those hours go. Training is one of the places they should go.
Write down three habits: feedback, paired work, and review time.
The fix is not a slogan about "investing in people." It is a short operating plan that still happens in a week when everyone is busy.
Schedule feedback, on a cadence, not only when something breaks. Put a recurring session on the calendar for each junior and the person responsible for their work, even if it is 25 minutes. Talk about a real piece of work, not a generic check-in. If the only feedback a new hire gets is after a mistake, they will wait to be told they were wrong instead of learning the standard in advance.
Pair people on real tasks. Once or twice a week, have a junior and a more experienced colleague work the same problem on a shared screen or a video call. The senior should think out loud. That is the remote substitute for sitting nearby, and a way to use AI in the room with someone who can catch a confident wrong answer.
Protect time for review. If AI is saving hours on drafting, reserve some of those hours for seniors to review junior work, including work a model helped produce. Write the reservation into the week the way you would write a client meeting. If every saved hour is filled with more output, the review that trains the next person is the first thing to disappear.
None of this requires an office five days a week. It does require a named owner, a written cadence, and scarce senior time spent on teaching. If you cannot name who is responsible for a junior's development, and when that work happens, the team will keep hiring for experience and wondering why the entry-level path got narrower.
The Stripe research will be debated, and it should be. The authors asked readers not to treat their model as proof. What does not depend on settling the AI-versus-remote debate is the management job: if informal training is thinner, formal training has to be thicker. Remote teams do not train juniors by accident. Do it on purpose.
We can help you with that.
NTx AI Consulting works with owner-led and partner-led privately held businesses that are currently using AI tools in everyday workflows, or would like to be. The complimentary consultation is a working session, not a product demo. If a deeper engagement makes sense, Phase 1 is an AI Disruption Readiness Audit. You can learn more at https://ntx.ventures/ai-consulting/.
Sources
- Ernie Tedeschi, "AI, work from home, and the young-graduate job market," Stripe Economics, Nov 12, 2026. https://www.stripeeconomics.com/p/ai-work-from-home-and-the-young-graduate
- Ozimek and Goldschlag (2026), as cited by Tedeschi.
- Emanuel, Harrington, and Pallais (2026); Bloom, Dahl, and Rooth (2026); Burn, Gentry, and Lahey (2026), as cited by Tedeschi.
- NTx Ventures, "Your team stopped asking each other questions. That's a problem." https://ntx.ventures/news/team-stopped-asking-questions.html
- NTx Ventures, "AI gave your team back a third of the time on the tasks it took." https://ntx.ventures/news/ai-freed-hours.html