Remote job postings require 25% more skills than similar in-person roles, study finds

The arithmetic behind that 64% is smaller than the number sounds. It is also more useful. Unemployment among college graduates under 29 went from 3.1% in 2017-2019 to 3.7% in 2022-2025 ^[2], which is 0.6 of a percentage point, or about 19% on the base ^[11]; assign the New York Fed's 64% ^[5] and roughly 0.38 points sit with working from home, leaving about 0.22 points for automation, cohort size, credential inflation and everything else the labour market did in those six years ^[12]. Nobody should wave that residual away. The comparison that carries more weight is the other one: experienced graduates moved the opposite way, 1.9% to 1.8% ^[3], so the spread between the two cohorts widened from 1.2 points to 1.9 points, up about 58% ^[13]. Demand shocks do not usually sort themselves that neatly by tenure.

What makes this a pipeline story rather than a hiring-market story is the posting-level finding. Holding title, employer and year constant, the remote version of a job asked for 25% more skills, more experience and slightly higher credentials than the in-office version ^[4], across more than 50 million ads in 28 European countries plus interviews and an experiment with US hiring managers ^[10]. That is a training cost moving to the other side of the table. Shinan Wang, the Kellogg co-author, puts the mechanism plainly: juniors need mentors, need to watch how others work, and firms respond by wanting people who already have the skills ^[6]. The effect concentrated in fully remote roles, with hybrid postings resembling in-person ones ^[7], which means the cheapest available remedy is a schedule rather than a budget line, and 35% of employees doing at least some work from home ^[1] is the size of the exposure.

Raising the entry bar has a name. A firm that does it stops funding the apprenticeship that historically produced future managers who were cheap to train and easy to move up ^[14], and starts renting three years of someone else's training at market price, and repeating that choice for a decade produces a bench nobody built. Meanwhile nearly half of recent graduates say AI is already affecting hiring in their field ^[8], which is a survey of belief, not of payroll.

This is probably wrong in one direction, but the more interesting version of the thesis is that the AI attribution is doing public-relations work for the remote-work attribution, because "the model took the job" describes an external force while "we stopped teaching" describes an internal cost that somebody approved. Emma Harrington, co-author of the Fed analysis, allows it may not be either-or ^[9]. Two ways this reads differently: if office mandates keep spreading and the 1.9-point cohort gap does not close, training was not the binding constraint and composition or automation wins the argument; and if the skills premium shows up in postings for experienced roles too, the 25% is screening noise rather than a missing classroom.