A carpenter's apprentice watched the master's hands. A junior analyst, until recently, got the white-collar equivalent for free: overheard negotiations, watched a senior colleague handle an angry client, absorbed a thousand small moves nobody ever wrote down. The first fully remote generation of workers gets neither. They are asked to learn a job they never see anyone do.
The research on remote work suggests this is the one place where the arrangement's costs are real and specific. The earliest rigorous evidence already contained the warning. In a nine-month randomized experiment at the Chinese travel firm Ctrip, employees who worked from home full time were 13 percent more productive and more satisfied, and quit less, but their rate of promotion conditional on performance fell (Bloom, Liang, Roberts and Ying, 2015). Output rose while careers slowed. Whatever promotion committees were rewarding, it was partly something transmitted in person, visibility, sponsorship, the informal apprenticeship, and the remote workers were locked out of it even as they outperformed.
The dose seems to matter. A later randomized trial at Trip.com with 1,612 university-educated professionals found that hybrid work, two home days a week, cut voluntary quits by a third (7.2 percent to 4.8 percent) with no detectable difference in performance reviews or in promotions over the following two years (Bloom, Han and Liang, 2024). The two studies cover different jobs and eras and are not a clean comparison, but the pattern they sketch is plausible: partial presence preserved the learning and visibility channel that full-time distance had cut. For a first-jobber weighing offers, the distinction between hybrid and fully remote is not lifestyle detail. It is the difference between an apprenticeship on reduced hours and no apprenticeship at all.
Documentation is the prosthetic
Fully remote work can still teach, but only in companies that rebuilt the transmission mechanism deliberately. The best-documented case is GitLab, which runs entirely remote on a handbook-first rule: decisions and processes go into a public handbook that would exceed 2,000 printed pages, on the stated logic that asynchronous work is impossible without strong documentation, because otherwise every follow-up question blocks someone for hours. These are the company's own claims, not measured outcomes, but the design principle travels: in a documented company, the newcomer can read what veterans know; in an undocumented remote company, that knowledge is locked in heads the newcomer cannot casually be near.
That yields the single best interview question a remote-curious graduate can ask: show me where things are written down. A real handbook, searchable decision logs, recorded walkthroughs of actual work: these are the remote equivalents of a desk next to the master. Their absence means the company is running on osmosis while employing people who cannot osmose.
Replace what the office did, piece by piece
The office taught through three channels, and each has a deliberate substitute the junior worker can request or build.
Observation: ask seniors to record themselves doing real work, a sales call, a code review, an editing pass, or to let you sit in silently on calls. Watching real work beats any training module, and recording costs the expert nothing extra.
Feedback: proximity generated constant micro-corrections; distance generates silence. The substitute is scheduled, explicit review of actual work product, asked for with a specific question rather than a general "any feedback?"
Visibility: the Ctrip result shows performance alone did not carry careers. The remote substitute is working in public: written weekly summaries, demos, documented decisions. This is not self-promotion; it is making the work observable in a medium that travels.
Measure the pump, not the hours
There is an older precedent for evaluating people without watching them. As Ricardo Semler has told it, after his collapse at 25 Semco stopped counting hours and started measuring whether the pump was delivered; absenteeism fell. Output-based evaluation, invented for a factory, is exactly what makes remote learnable: when the deliverable is the unit of judgment, a junior's work can be seen, assessed, and improved without anyone sharing a room.
The remote apprenticeship problem is real, and pretending presence never mattered has cost part of a generation its informal education. But the answer is not nostalgia for offices. It is choosing, and pushing employers toward, the structures that made presence unnecessary: documents instead of proximity, explicit feedback instead of ambient correction, visible output instead of visible hours. Apprenticeship was never about the room. It was about access to how the work is actually done, and access can be engineered.
Sources
- Bloom, N., Liang, J., Roberts, J. & Ying, Z. J. (2015). Does Working from Home Work? Evidence from a Chinese Experiment. Quarterly Journal of Economics 130. RCT: full-time WFH raised performance 13% and satisfaction, but promotion conditional on performance fell, an early measure of remote work's career cost. https://academic.oup.com/qje/article-abstract/130/1/165/2337855
- Bloom, N., Han, R. & Liang, J. (2024). Hybrid working from home improves retention without damaging performance. Nature 630. RCT with 1,612 professionals: two home days cut quits by one third (7.2% to 4.8%) with no difference in performance or promotions over two years. https://www.nature.com/articles/s41586-024-07500-2
- GitLab (accessed 2026). Handbook-first and asynchronous work. GitLab all-remote handbook. Company's own case: a public handbook exceeding 2,000 printed pages as the operating base, on the claim that async work is impossible without strong documentation. https://handbook.gitlab.com/handbook/company/culture/all-remote/asynchronous/