Should we hire fewer juniors?

The evidence says many firms already are, and does not show that doing so pays. A Harvard résumé study finds junior employment falling at firms that adopt generative AI while senior employment holds, driven by slower hiring. No study read for this page measures what the saving costs later. Two older studies suggest where the cost would appear: experience that is not kept up decays quickly inside an organisation, and part of a skilled person's performance belongs to the place they learned it, so seniors cannot simply be bought in when the juniors have gone. The decision turns on whether the junior work is still needed to train anyone.

Firms that adopt generative AI are hiring fewer juniors#

Hosseini Maasoum and Lichtinger of Harvard use résumé data on 65 million workers at more than 280,000 firms and identify adoption from job postings for people hired to integrate generative AI. Following adoption, they report, "junior employment declines in adopting firms relative to non-adopters, while senior employment trends remain largely unchanged", concentrated in the occupations most exposed to AI and "driven primarily by slower hiring rather than increased separations". GenAI-exposed tasks also become less likely to appear in junior task bundles. It is an unrefereed working paper and the authors call the link an association. The sector-wide picture, from Stanford's payroll study, is a 19 per cent gap for 22 to 25 year olds in exposed occupations that opens through reduced hiring, and the case that the whole pattern is about remote work and not AI is set out on is it harder to get a first job now.

Experience that is not renewed can disappear within a year#

Anelí Bongers, in a 2017 PLOS ONE study of flyaway costs for three US fighter aircraft, estimated how much accumulated production experience carried over from one year to the next. For the F-22A and F-35A the parameter was "not significantly different from zero", which she reads as meaning that "the depreciation of experience in the production of these units is total, on an annual basis." It is a study of aircraft production lines, estimated from cost data, and nobody has shown the same for professional judgement. It matters here as a measured case of an organisation losing what it had learned faster than its managers assumed.

Seniors are partly made inside the firm#

Huckman and Pisano, in Management Science in 2006, examined cardiac surgeons who operated at several hospitals. Using patient mortality as the outcome, they found a surgeon's quality at a given hospital improved significantly with recent volume at that hospital and not significantly with volume elsewhere, and concluded that "surgeon performance is not fully portable across hospitals". Their preliminary explanation is familiarity with the organisation's assets. For a hiring decision, the implication is that a mid-career recruit from outside brings part of what an internally developed person has, and not all of it.

What none of these studies measured#

• The later cost of fewer juniors. The Harvard study measures hiring and employment, and does not follow what happens to seniority pipelines.
• Whether the lost junior work trained anyone. The Harvard data records tasks leaving junior roles. Nothing in it tests whether those tasks were how people learned.
• Professional work. The aircraft and surgery studies sit in settings with measurable output, and neither concerns knowledge work with AI.
• The counterfactual. Whether firms that kept hiring juniors did better is unmeasured in what was read here.

The saving is booked this year and the bill arrives in the senior tier#

This section is interpretation, kept apart from the evidence above.

The argument from The Missing Rungs (21 September 2025) is that the entry-level tasks were how people advanced, so removing them removes the route and not only the headcount. The two older studies make that argument measurable in principle: if experience decays fast and performance is partly local, a firm that stops developing juniors depletes the stock its future seniors come from, and the market cannot fully refill it. That inference joins studies of different settings and is untested. The question for a leader is therefore concrete. For each junior task AI now does, who will learn it, and by what other route? The routes available are discussed on how do juniors become senior and how do you keep expertise in an organisation.

Questions to settle before cutting an intake#

• List the tasks the juniors did and mark which ones taught judgement. Automate the ones that did not and keep a supervised share of the ones that did.
• Name who will hold the senior roles in five years. If the answer is external hiring, price the part of performance that does not travel.
• Keep a smaller, deliberate intake before dropping it to zero. A restart after a gap loses the people who would have trained the next cohort. This is a judgement and no study here tests it.
• Record the decision and its assumption. If you cut because AI covers the work, note which task and what quality, so the decision can be tested later.

Key sources

• Hosseini Maasoum, S. M. and Lichtinger, G. (2026). Generative AI as Seniority-Biased Technological Change. SSRN working paper. Graded entry.
• Brynjolfsson, E., Chandar, B. and Chen, R. (2026). Canaries in the Coal Mine? Stanford Digital Economy Lab. Graded entry.
• Bongers, A. (2017). Learning and forgetting in the jet fighter aircraft industry. PLOS ONE, 28 September 2017. Graded entry.
• Huckman, R. S. and Pisano, G. P. (2006). The Firm Specificity of Individual Performance: Evidence from Cardiac Surgery. Management Science 52(4), 473 to 488. Graded entry.

Related SuperSkills research#

On the job seeker's side, is it harder to get a first job now. On the pipeline, the missing rungs. On whether the roles disappear, will AI replace entry-level jobs. On hiring for the capability that remains, how do you hire for judgement.

About this research#

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. The Harvard abstract, the Stanford page, the PLOS ONE abstract and the Management Science abstract were read at source on 3 October 2026, and all four are graded in the evidence base. Firm-specific performance is an established term and not a SuperSkills coinage.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.