Is it really AI that's killing junior hiring – or is remote work the hidden culprit?

The story has been told so many times it has hardened into consensus. Generative AI is hollowing out the entry-level job market. It is doing the work that graduates and junior associates once performed – the data wrangling, the draft documents, the first-pass analysis – and firms no longer need to hire humans to do it. For Australian HR professionals, this story has been front-of-mind for the past two years.

One-third of Australian organisations have already slowed hiring for entry-level positions, and 60% expect to reduce such recruitment within three years, according to research from IDC commissioned by Deel. CPA Australia's Business Technology Report 2025 found that eight per cent of Australian businesses have reduced hiring of junior or entry-level accounting and finance staff — behind the Asia-Pacific average of 17%, and well below the 32% seen in Mainland China. The dominant explanation for these numbers is always the same word: AI.

Read next: Achieving productivity in flexible workplaces

But 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 – suggests that Australian HR teams may be misreading the situation in a way that will have serious long-term consequences. The real culprit, the authors argue, may not be artificial intelligence at all. It may be the flexible work arrangements those same organisations fought so hard to keep.

The paper, titled 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, spanning 2017 to 2025.

Its central finding is disarmingly simple. When researchers measure the effect of generative AI exposure and working-from-home exposure on junior hiring separately, both look like powerful explanatory factors. A two-standard-deviation increase in either exposure predicts roughly a 4–5 percentage point fall in the junior share of new hires by 2025. Both tell a convincing story on their own.

Read next: Employee who insisted on WFH loses job for failing to return to office

The trouble is that they are measuring largely the same thing. The WFH exposure measure and the GenAI exposure index used in the study have a Spearman rank correlation of 0.77 across 683 occupations. Software developers, accountants, and management consultants sit near the top of both rankings. Electricians, janitors, and construction workers sit near the bottom. The occupations most exposed to AI automation are, almost without exception, also the occupations most amenable to remote work.

When the researchers put both variables into the model simultaneously, the picture shifts dramatically. The WFH effect holds firm. The AI effect collapses – often becoming statistically indistinguishable from zero. This is not a minor methodological footnote. It is a fundamental challenge to the prevailing explanation for what has happened to early-career hiring across the English-speaking world.

The junior hiring decline is real, and Australia is squarely inside it. Entry-level hiring in Australia has contracted sharply, mirroring a global pattern. According to Randstad's 2025 Global Workplace Blueprint, organisations worldwide are pulling back on graduate and early-career recruitment. In Australia, Fifth Quadrant's SME Sentiment Tracker showed just 9% of small and medium enterprises were actively hiring in December 2025, down from 19% in October of the same year.

According to the Australian Bureau of Statistics, 36% of employed people usually worked from home in August 2024 – down slightly from 37% in 2023, but still well above pre-pandemic levels. The Lambert-Schindler paper's Australian event-study data tells a pointed story: junior hiring in Australia tracked the other three countries almost precisely, falling roughly 8–10 percentage points below the 2019 baseline by 2025, with the sharpest descent beginning in late 2022 – coinciding not only with the ChatGPT release but with the crystallisation of pandemic-era remote work into permanent organisational policy.

Read next: CEOs oppose plan to legislate WFH

That coincidence of timing is the crux of the problem. When HR professionals, CFOs, and chief executives point to ChatGPT's November 2022 launch as the forcing event for their reduced junior intake, they may in fact be pointing to the moment when temporary WFH arrangements became permanent. The two events are nearly inseparable in the data, and no single-variable analysis can reliably separate them.

The paper's theoretical framework is rigorous, but its intuition is familiar to any experienced HR leader. Firms hire junior workers not merely for the work they produce today, but for the workforce they will become. That investment makes sense when the costs of turning a graduate into a capable professional are manageable – when a junior can absorb knowledge through proximity, receive informal feedback in the corridor, and gradually earn greater autonomy through visible performance.

Remote and hybrid arrangements raise the cost of all of that. Supervision takes more deliberate effort. Informal feedback loops break down. The learning that happens by sitting next to a senior colleague – watching how they handle a difficult conversation, how they structure a board paper, how they navigate organisational politics – does not transfer across a video call with the same fidelity.

Research cited in the paper is consistent on this point. Studies of collaboration networks found that firm-wide remote work made those networks more siloed. Separate research on young software engineers found that in-person office time delivered the largest productivity benefits for the youngest workers. A study of a fully remote firm found that in-person onboarding raised later productivity and reduced attrition, even when employees subsequently returned to remote work. Research published in The Quarterly Journal of Economics in 2026 found that proximity to co-workers significantly increases feedback among junior employees, with the largest gains for younger workers.

These are not arguments against flexible work. They are arguments that flexible work arrangements, when applied without adjustment to how early-career talent is developed, make the return on hiring junior staff harder to realise – and that firms respond to that harder calculation by hiring fewer of them.

The model in the paper formalises this mechanism: as supervision costs rise and on-the-job learning rates fall, the rational firm tilts its hiring toward more senior workers who have already demonstrated their capabilities and who require less organisational scaffolding. That tilt, multiplied across thousands of firms, produces the aggregate hiring decline now being observed in the data.

It would be a mistake to read this paper as an exoneration of generative AI. The authors are explicit on this point. Their findings relate specifically to occupation-level exposure designs – the standard method by which researchers infer AI's effect on hiring – and to the period up to 2025. They do not claim that AI has had no labour market effects.

Research cited by HRD Australia found that employment for young software developers declined by nearly 20% by July 2025, with similar drops for computer occupations and service clerks more broadly. The IDC/Deel research found that 93% of Australian organisations are anticipating a reduction in the general hiring of entry-level staff in the next one to five years, and 74% reported fewer on-the-job development opportunities for junior employees. In Australia, 91% of organisations report that roles have already changed or disappeared due to AI adoption in the past year alone.

What the paper challenges is the causal attribution. When a firm reports that it is hiring fewer graduates because of AI, the question Lambert and Schindler would ask is: could the same firm's hybrid policy be doing at least as much of the explanatory work?

The robustness of their finding is notable. Across alternative exposure measures, alternative outcome definitions, leave-one-out occupation and country tests, non-parametric controls, and measurement-error simulations, the WFH effect holds. The AI effect does not. A confounder explaining only 1–3% of residual variation in both treatment and outcome would be enough to eliminate the AI coefficient – eliminating the WFH coefficient would require a confounder five times as powerful.

The paper also presents a particularly striking further test. When the researchers replace predicted AI and WFH exposure with actual observed WFH adoption – tracking which firms and regions explicitly offered remote and hybrid roles in 2021–22 – they still find a robust negative effect on junior hiring in the years that followed. The causal story survives contact with real-world adoption data.

Indeed's Hiring Lab data from April 2026 complicates the AI narrative further. Australian graduate postings in highly AI-exposed occupations fell sharply through 2025, but rose in early 2026 compared to a year earlier – suggesting healthy forward demand for the current year. If AI substitution were the primary structural driver, that recovery would be difficult to explain.

The Australian HR Institute's own research shows a more nuanced picture than public discourse suggests. Four in 10 respondents said they are seeing an increase in entry-level roles due to AI, while 19% report a decrease. Public-sector organisations — which generally maintained more structured development frameworks and had less flexibility to implement sweeping hybrid arrangements without formal policy changes — reported stronger growth in entry-level opportunities. That divergence may say something important about organisational design rather than technology.

Meanwhile, more than 68% of Australian respondents now rate specific AI tool skills and coding certifications when looking for entry-level staff – a finding that suggests the nature of junior roles is evolving, not simply disappearing. That is consistent with Lambert and Schindler's broader point: AI is reshaping what junior workers do, but it may be WFH that is discouraging firms from hiring junior workers at all.

The policy implications, as the paper frames them, are genuinely more optimistic than the AI-displacement story would suggest – if organisations are prepared to act on them.

As HRD has reported, CFOs are simultaneously intensifying succession planning and reducing junior hiring – two intentions that are in direct tension. The senior leaders of 2035 are the junior hires of today, and today, those hires are not being made. The Lambert-Schindler paper suggests this problem may have a more tractable cause than technology substitution: it may be a management practice problem, not an inevitable consequence of automation.

If AI were the primary driver, the remedies would have to be correspondingly blunt – wage subsidies, differential taxation, structural industry policy. If WFH is the primary driver, the remedies lie within the reach of HR functions: deliberate onboarding design, structured mentoring, purposeful in-person time for early-career cohorts, rotation programmes, and hybrid policies that apply differently to workers in their first two years than to experienced staff who have already built the relational and contextual knowledge to work effectively at a distance.

There is a longer-run risk in the current trajectory that the prevailing AI narrative obscures. Early-career jobs are the mechanism by which general human capital is accumulated. Workers who miss the window at the bottom of the ladder – who cannot find the entry-level role that would give them firm-specific knowledge, professional networks, and the tacit workplace skills that cannot be taught in a training course – do not simply acquire those things later. Research on labour market cohorts consistently finds scarring effects: workers who enter the market during periods of restricted junior hiring earn less and advance more slowly, sometimes for the rest of their careers.

The aggregate result, if current trends continue, is a thinning of the experienced talent pool a decade from now – at exactly the moment when organisations will need capable professionals to supervise and direct the AI systems they are currently deploying. That is the paradox at the heart of the junior hiring decline, and it is one that no amount of AI investment will resolve if the underlying management problem is not first identified and addressed.

For Australian HR leaders, the most useful immediate step is diagnostic. Audit whether your organisation's junior hiring decline correlates more closely with the teams and business units that went hybrid earliest, or with the ones deploying AI most aggressively. The answer may provide some interesting clarity.

Examine whether your onboarding and mentoring infrastructure has been rebuilt to compensate for the proximity losses of hybrid work – or whether it still implicitly assumes that junior staff will absorb knowledge through informal daily contact that no longer exists in the same form. There are practical ways to help young staff develop professional skills in changing workplace environments, including structured mentoring, deliberate exposure to senior decision-making, and investment in learning pathways that do not depend on physical co-location.

Resist, above all, the temptation to treat early-career headcount reduction as an immovable consequence of technology – rather than as a management choice that can be revisited. The broken ladder Lambert and Schindler describe is a policy problem. If the ladder is broken in large part because of how firms chose to reorganise work after a once-in-a-century pandemic, then firms retain the agency to rebuild it.

That is a more demanding conclusion than blaming AI. It is also a far more useful one.