Enter AI-exposure on its own and it predicts roughly a 5-percentage-point fall in the junior share of new hires by 2025. Enter remote-work exposure on its own and you get the same 5-point drop. Enter both together and one of them collapses: the AI coefficient "attenuates sharply and often becomes statistically indistinguishable from zero," while remote work stays "a strong and robust predictor of declining junior-share across every specification" (Lambert & Schindler, 2026). That is the finding most mid-market hiring plans are built to contradict.
If you run operations at a 50–500 FTE company, you have almost certainly heard — and possibly repeated — the story that generative AI is eating the entry-level job. It is the story behind the hiring freeze on junior roles, the deferred graduate cohort, the quiet decision to "let AI handle the work a first-year would have done." A new working paper analyzing 243 million hires says that story is measuring the wrong variable. The early-career pipeline is collapsing, but the force doing the damage is one your AI strategy can't touch — and your real-estate and onboarding policy can.
The Misdiagnosis Sitting in Most Hiring Plans
The dominant 2026 narrative is clean and intuitive: large language models are good at exactly the bounded, well-specified tasks junior staff used to cut their teeth on, so the rational firm automates those tasks and stops hiring the juniors. The numbers that get quoted alongside it are real — early-career hiring has genuinely fallen. Across the United States, United Kingdom, Canada, and Australia, the share of new positions filled by early-career workers has dropped 8–11 percentage points below pre-pandemic levels (Innovative Human Capital, 2026). The contraction is not imagined.
The problem is the attribution. The occupations most exposed to generative AI — white-collar, routine-cognitive, desk-bound roles — are very nearly the same occupations that went remote after 2020. When two forces move together, a model that only looks at one of them will hand that one the credit for both. Most hiring plans have done exactly that: they have read a correlation between AI exposure and shrinking junior intake and concluded that AI is the cause. The policy that follows — slow the graduate program, lean on AI to cover entry-level output — treats the symptom of a variable the analysis never isolated.
What the 243-Million-Hire Paper Actually Found
Peter John Lambert and Yannick Schindler set out to separate the two. Their May 2026 working paper, The Broken Ladder: AI, Remote Work, and Early-Career Hiring, draws on 243 million new hires and 407 million online job postings across the US, UK, Canada, and Australia from 2017 to 2025 — a sample large enough to estimate AI-exposure and work-from-home exposure as distinct effects rather than a single blurred trend (Lambert & Schindler, 2026).
Tested in isolation, each force looks like a culprit. A two-standard-deviation increase in either generative-AI exposure or remote-work exposure predicts roughly a 5-percentage-point fall in the junior share of new hires by 2025, alongside about a 3-percentage-point drop in the share of job ads requiring only limited experience. On its own, AI exposure is a statistically respectable suspect.
It does not survive the lineup. When both variables enter the same specification, the AI coefficient attenuates sharply toward zero and frequently loses statistical significance, while the work-from-home coefficient holds steady across every specification the authors run. Their blunt summary: a simple binary remote-work indicator is "enough to render the GenAI effect insignificant." The AI signal, in other words, was largely remote work wearing an AI costume. The correlation was real; the causation was misassigned.
This is the kind of result that should change a decision, not just a slide. If the entry-level decline were AI-driven, the operational response — automate junior tasks, hire fewer juniors — would at least be internally consistent. Because it is remote-driven, that same response does nothing to the actual mechanism and quietly forecloses a pipeline the company will need in three years.
The Mechanism: Mentorship Doesn't Travel Well Over Slack
Why would remote work, specifically, compress junior hiring? The honest answer is that a junior hire is only worth the cost if someone can develop them, and development is disproportionately a co-located activity. The evidence here predates the AI panic and is unusually concrete.
In a study of software engineers, Natalia Emanuel, Emma Harrington, and Amanda Pallais found that engineers sitting in the same building as their teammates received 22 percent more comments on their code than engineers split across buildings — and that this feedback flowed mainly to junior engineers from senior ones (Emanuel, Harrington & Pallais, NBER, 2024). Proximity didn't just add a few in-person chats on top of the same online feedback; engineers who lost physical proximity exchanged less feedback online as well. Face-to-face contact and digital communication turned out to be complements, not substitutes (Federal Reserve Bank of New York, 2024).
The trade-off the same study documents is precisely the one a hiring manager has to price: proximity raises long-run human-capital development at the cost of short-run output, dampening immediate pay raises but boosting them over a career. Co-location is an investment in the junior worker that pays off later. Strip out the proximity and you strip out the return — at which point the junior hire stops penciling out, and the firm rationally stops making it. Remote work didn't make juniors less capable. It made them more expensive to develop and less obviously worth developing. That is the lever behind the broken ladder, and it has nothing to do with what your AI tools can or can't do.
The Counter-Argument: "It's Both, and AI Is Still Coming"
The sharpest objection from an experienced operator is fair, and it deserves a direct answer. AI is still genuinely capable of junior-level work, the technology is improving fast, and a single working paper — however large — shouldn't override the structural logic that automation displaces the most automatable tasks first. Isn't "it's remote work, not AI" just a comforting story with a longer one coming behind it?
Two things are true at once. First, the paper does not claim AI has zero effect on labor markets; it claims that AI exposure does not independently explain the junior-share decline observed through 2025 once remote work is controlled for. That is a precise, bounded finding about one outcome over one window — not a forecast that AI will never reshape early-career work. A leader can take the econometrics seriously and still plan for AI's trajectory.
Second, and more useful: even if you believe AI pressure on junior roles is coming, the WFH finding tells you the lever you control right now is the wrong one to be pulling on the question of pipeline. Pausing or slowing your AI rollout will not re-open the junior funnel, because AI is not what closed it. You would be paying a real cost — falling behind on the productivity gains AI does deliver — to fix a problem AI did not cause. The disciplined move is to stop treating AI rollout pace and junior-pipeline health as the same dial. They are different dials, and the paper just told you which one is wired to the outcome.
The Q3 Move: Stop Adjusting the AI Dial, Start Scheduling Proximity
The correction is not a hiring philosophy; it is a calendar. If remote work collapses the junior pipeline by starving development of the in-person mentorship it depends on, then the operational fix is to engineer that proximity back in for the people and the window where it matters most — and the econometrics point at a specific window: early tenure.
Three moves are installable this quarter. First, treat the first six months of a junior hire's tenure as a deliberately higher-density in-person period. Not a blanket return-to-office mandate — a targeted one, where the hybrid-policy days you schedule are concentrated around new early-career staff and the senior people who will actually review their work. The proximity research is specific that the feedback flows from experienced seniors to juniors; co-locating juniors with each other accomplishes little.
Second, build cohort tracks. Hiring juniors one at a time into remote-default teams maximizes isolation. Hiring them in cohorts with a structured, in-person onboarding spine creates the peer density and the visible mentorship that make development happen — and makes the junior hire pencil out again.
Third — and this is where most plans go wrong — base who gets that high-touch onboarding on something other than the org chart. Tenure, role, and stage of development are individual attributes, not job-title attributes; the juniors who most need dense early mentorship are not always the ones a generic policy would flag. That is a measurement question, and measurement questions are answered better with data than with intuition. Scovai's assessment base is built to profile exactly the developmental and role-readiness traits that tell you which early-career hires need the structured in-person track and which can thrive in a lighter-touch hybrid setup — so the proximity you can afford to schedule lands on the people whose long-run trajectory it actually changes.
Lambert and Schindler handed operations leaders an uncomfortably clean receipt: the force collapsing your junior pipeline is the one in your hybrid-work policy, not the one in your AI stack. The single decision this leaves on your desk this quarter is narrow. Pull up your early-career hiring plan and find the lever it pulls to "fix" the funnel. If that lever is labeled AI — pausing it, leaning on it to cover entry-level work, waiting for it to settle — you are adjusting a dial that isn't connected to the outcome. The one that is connected is the in-person schedule for your first-year hires, and you can change it before the quarter ends.
