The real reason junior hiring is collapsing may not be AI. It may be your remote work policy

Across 243 million new hire records and four countries, the evidence points to flexible work — not generative AI — as the more robust predictor of declining early-career investment.

The statistics are stark, and they have been getting starker for three years. Entry-level software engineering postings in the United States dropped sharply between 2023 and 2024, according to Stanford's Digital Economy Lab. The share of tech jobs requiring three years of experience or less fell from 43% in 2018 to 28% in 2024. Employment for young software developers aged 22 to 25 declined nearly 20% from its late-2022 peak. In the broader labour market, job losses linked to AI are concentrated at junior and mid-level positions, with the highest reported impact at entry level — 63% of organisations report this effect.

The explanation that has taken hold across corporate America is as intuitive as it is consequential: generative AI is doing the work that junior hires once did, and organisations no longer need as many people at the bottom of the career ladder. The risk of entry-level pathways narrowing is already influencing what young people study and which careers they pursue.

A major new working paper — published in May 2026 by researchers at the University of Warwick, the London School of Economics, and Oxford's Ellison Institute of Technology — does not dispute that AI is reshaping the workplace. What it argues, with considerable empirical force, is that the standard evidence for AI's causal role in suppressing junior hiring may be confounded by another factor that has been hiding in plain sight: the permanent post-pandemic shift to hybrid and remote work.

The paper, The Broken Ladder: AI, Remote Work, and Early-Career Hiring, by Peter John Lambert and Yannick Schindler, draws on 243 million new hire records and 407 million online job postings across the United States, the United Kingdom, Canada, and Australia between 2017 and 2025.

When AI exposure and WFH exposure are measured in isolation, both appear to powerfully predict declines in junior hiring. A two-standard-deviation increase in either predicts roughly a 4–5 percentage point fall in the junior share of new hires by 2025, with no evidence of pre-trends before 2022.

The reason both effects look similar is that they are tracking the same occupations. The WFH exposure index and the GenAI exposure index have a Spearman rank correlation of 0.77 across 683 occupations. Software developers, data scientists, lawyers, and management consultants rank near the top of both. Electricians, construction workers, and truck drivers rank near the bottom. When an AI-only study uses occupation-level exposure variation to identify AI's effect, it is almost certainly picking up WFH effects operating on the same workers.

In joint models that include both exposures simultaneously, the WFH coefficient remains large, stable, and highly significant across every specification. The AI coefficient attenuates sharply — often to zero, and sometimes reverses sign. A confounder explaining only 1–3% of residual variation in both treatment and outcome would be enough to eliminate the AI coefficient entirely. Eliminating the WFH coefficient would require a confounder five times more powerful.

The authors also conduct a direct test using actual WFH adoption — tracking which firms explicitly offered remote or hybrid work in 2021–22, matched against their junior hiring in 2023–25. The causal story survives contact with real-world adoption data.

The United States entered the post-pandemic period as one of the world's highest WFH-adoption economies, and that position has not materially changed. Gallup's most recent data shows hybrid as the dominant arrangement for remote-capable workers, with 52% in hybrid arrangements and 26% exclusively remote. The share of paid workdays spent at home was approximately 27% in July 2025, according to the Survey of Working Arrangements and Attitudes — multiples above any pre-pandemic baseline.

The return-to-office movement has been loud, but the data suggests it has been limited in effect. The Owl Labs State of Hybrid Work 2025 found that flexible working hours are the most important factor in working lives for 83% of employees, surpassing even location flexibility, and that not allowing flexible hours was the top reason employees would decline a job offer. The structural conditions that the Lambert-Schindler paper identifies as suppressing junior hiring are, in most American knowledge-sector organisations, deeply entrenched.

It is worth noting the distributional dimension. Research on remote work availability shows that senior-level roles have the highest rates of flexibility, with nearly half including some form of location flexibility. Entry-level positions show the most limited options — only 28% of entry-level jobs offer any flexibility. This creates an asymmetry: senior workers get both the benefits of WFH and the development infrastructure built up over years in physical offices. Junior workers enter an environment where WFH norms were established before they arrived — and where the informal learning mechanisms those norms disrupted were designed for, and validated by, an in-person workplace they never experienced.

The paper's theoretical contribution is to show that WFH suppresses junior hiring not by replacing junior workers with remote tools, but by raising the cost and reducing the return of investing in early-career talent. Firms hire juniors not just for what they produce in year one, but for the experienced professionals they will become in years three through ten. That investment depends on knowledge-transfer mechanisms that proximity enables and distance disrupts: informal mentoring, incidental observation of senior decision-making, feedback loops that allow rapid skill accumulation in the early career years.

When those mechanisms erode, the return on junior investment falls. When the return falls, rational firms tilt their hiring toward workers who have already built their capabilities elsewhere — under prior conditions, at other firms — and who need less organisational scaffolding to be productive. Replicated across thousands of American organisations, that tilt produces the aggregate junior hiring decline now being attributed to AI.

This mechanism is not speculative. Research cited in the paper found that firm-wide remote work made collaboration networks more siloed. A study of female software engineers found that proximity to colleagues significantly increased feedback frequency and quality, with the largest gains for younger workers. A study of a fully remote firm found that in-person onboarding raised later productivity and reduced attrition. Owl Labs' own data found that 25% of workers worry junior staff are losing the unplanned learning that occurs through physical co-presence with senior colleagues — and that 51% of workers identify mentoring as better suited to in-office time.

Stanford's study confirms that AI tools have an influence on entry-level jobs, but the impact is not across the board. Occupations with the highest estimated automation shares have experienced declining employment for the youngest workers — but occupations with the highest augmentation shares did not report a similar pattern.

Gartner's October 2025 survey of customer service and support leaders found that only 20% had actually reduced agent staffing because of AI, despite AI being cited in tens of thousands of publicly announced layoff plans. The gap between stated rationale and measurable staffing effect is not trivial — and it is precisely the kind of gap that a WFH-confounding problem would produce.

IBM's decision announced in 2026 to triple US entry-level hiring — explicitly describing early-career investment as valuable precisely because junior roles have already been redesigned around AI augmentation — offers a data point the AI-displacement thesis struggles to accommodate. So does the fact that the share of graduate job postings in AI-exposed occupations began recovering in early 2026 after two years of decline.

The Lambert-Schindler paper does not offer an argument against AI investment, flexible work, or hybrid arrangements. It offers an argument that organisations may be making a costly attribution error — and that the error is causing them to misidentify a management problem as a technology inevitability.

The immediate practical step is diagnostic. Does your organisation's junior hiring decline track most closely with the business units that adopted hybrid and remote work earliest — or with those deploying AI tools most aggressively? For most American knowledge-sector organisations, the answer to this question is likely more informative than any AI benchmarking exercise.

The second step is structural. Flexible working hours are now the most important factor in working lives for most American employees — a preference that cannot simply be overridden by return-to-office mandates. But that preference can be accommodated while designing deliberate compensatory infrastructure: structured mentoring programmes, explicit in-person time for early-career cohorts, rotation programmes that expose junior workers to diverse functions and senior relationships, onboarding frameworks rebuilt for distributed teams.

The third step is strategic. The senior leaders of 2035 are the junior hires of 2025 — and today, in large parts of the American knowledge economy, those hires are not being made. If the broken ladder is primarily a management challenge rather than a technological inevitability, the cost of not fixing it compounds every year that passes without intervention. The talent that is not developed does not disappear. It either does not enter the pipeline at all, or enters it at a competitor or a country that has designed its early-career infrastructure more thoughtfully.

American HR has a long track record of solving hard organisational problems. The junior hiring decline is one of them — if it is correctly diagnosed.