Distance is cheapest for the people who need it least
September 2026
This is the finding we like least and cite most.
Natalia Emanuel of the Federal Reserve Bank of New York, Emma Harrington of the University of Virginia and Amanda Pallais of Harvard published an analysis in June 2026 of who has been paying for distributed work. The answer is young people.
Between 2017–19 and 2022–25, unemployment among college graduates under 29 rose from 3.1 percent to 3.7 percent — an increase of about a fifth. Over the same period, unemployment among college graduates 29 and older fell slightly, from 1.9 to 1.8 percent. The labour market did not get worse. It got worse in one place.
The authors' estimate is that remote work can account for 64 percent of that increase. The mechanism is visible in firm-level data from a Fortune 500 company: workers seated near their colleagues received substantially more feedback and mentorship than workers who were not, and that feedback tapered sharply with distance. The loss fell hardest on the youngest employees, who have the most to learn from it. When offices closed, the firm shifted hiring toward experienced workers. When offices reopened, it resumed hiring juniors — except onto teams that stayed distributed, where it kept hiring experience instead.
Read that last sentence again, because it is the practical one. The firm was not hostile to junior hiring. It was responding rationally to an environment where it could no longer develop juniors cheaply. Distributed teams did not decide to stop training people. They stopped being able to, and then stopped hiring the people who needed training.
This is the strongest available argument against the position that remote work is costless, and we are not going to soften it. We would only add what it does not say. It does not say hybrid fails; the trial evidence on hybrid retention is good. It does not say offices are necessary. It says feedback is, and offices have been the default delivery mechanism for it. Distance is expensive because organisations let mentorship stay informal, ambient and unscheduled — a byproduct of seating rather than a practice.
If mentorship is ambient, distance destroys it. If it is designed, written down, assigned to named people and put on a calendar, distance is survivable. Almost nobody does the second thing, which is why the first result keeps showing up in the data.