What remote work took away

At one Fortune 500 online retailer, some software teams sat together in a single building and others were split across two buildings on the same headquarters campus, roughly a ten-minute walk apart.¹ Before the pandemic, engineers on the single-building teams received 23.9 percent more comments on their code than engineers on the split teams.¹ A ten-minute walk reduced feedback about as much as having teammates several states away.¹ When the offices closed and everyone went home, the gap largely disappeared.¹

That detail says something about what an office does. Nobody designed those two buildings to suppress code review. The walk simply made the easy thing, turning around and asking, slightly less easy, and that was enough. A workplace is full of mechanisms like this: contacts that happen because people are near each other, habits held in place by a room, movement that comes free with getting somewhere. None of them is in a job description. Remote work keeps the job description and removes some of the mechanisms.

This piece started from a broader version of that claim, that offices and commutes supplied the transitions that organized attention, and that remote work took them away. The evidence moved it. The losses that show up most clearly are not about transitions at all. They are about unplanned contact, and they fall on some people far more than others.

The contacts nobody scheduled

The clearest test comes from Microsoft. Before the pandemic, managers decided case by case who could work from home, and few did; in March 2020 the company began sending employees home, and by April all of its non-essential US staff were remote.² Researchers used the emails, calendars, messages, calls and working hours of 61,182 employees to compare people whose own situation and whose colleagues' situation changed by different amounts, which, under the study's assumptions, separates the effect of remote work from the rest of the pandemic.²

Remote work made the company's collaboration network more static and more siloed.² The number of ties that bridged otherwise separate parts of the network fell by 9 percent of its February level, and the share of collaboration time spent with those bridging ties fell by 41 percent.² People did not replace hallway contact with video calls; synchronous communication fell overall and email and instant messages rose.² The authors' conclusion is careful: these changes may make it harder to acquire and share new information across the network.²

Bridging ties are the contacts outside your own group, often weak, often unplanned. Nobody books a meeting with the person from another team they happened to stand next to. In an office those contacts are a by-product of sharing space. At home they have to be scheduled, and the Microsoft data suggest most of them simply were not.

Feedback you overhear

The retailer study is more specific about who loses. Researchers followed its engineers from 2019 to 2024, through the office closures and through return-to-office mandates in 2022 and 2023.¹ Sitting near teammates increased the feedback engineers received by 18.3 percent, and the comments that disappeared when teams were apart were disproportionately the helpful, actionable ones.¹ Being nearby made it easier to ask: co-located engineers asked more follow-up questions and drew feedback from a wider range of colleagues.¹

The gains went mostly to new hires and younger engineers, the people with the most to learn.¹ Women were also more affected, partly because they asked fewer coworkers for feedback when they could not ask in person.¹ After a three-day office mandate brought co-located teams back together, their engineers became less likely than engineers on distributed teams to write code that was later deleted or to introduce bugs, and less-tenured and younger engineers gained roughly twice as much as the average.¹

There is another side, and it is why the argument is not simply that offices are better. Feedback costs the person giving it. Experienced engineers, who give the most, wrote less code when they sat near their colleagues; the gap closed when offices shut and widened again when they reopened.¹ One way to read this is that the office was quietly taxing senior people's output to train junior people. Remote work lowers the tax and, with it, the training. When the offices closed, the firm shifted toward hiring older workers, buying experience it could no longer grow as easily.¹

The walk

Some of the machinery is physical. In a Finnish study, 99 hybrid workers wore thigh-mounted accelerometers on both office days and remote days, so each person was compared with themselves.³ On remote days they spent 45 more minutes sitting or lying down, and less time standing and in light or moderate-to-vigorous activity; they also slept about 15 minutes longer.³ The researchers connect part of the difference to the missing commute, and part to the fact that home rarely has a standing desk.³

Places also hold habits. When psychologists followed students transferring to a new university, habits such as exercising and reading the newspaper survived the move only when parts of the old context stayed the same, for instance reading the paper with other people.⁴ When the cues changed, behavior came under intentional control instead, and people did what they currently meant to do.⁴ That study was not about work, so applying it to the office is an inference. But it describes the same mechanism: a routine that lived in a building can fail to make the trip home.

The part people miss least

The commute is the obvious candidate for lost infrastructure, and the evidence for it is the weakest. A study of 1,736 commuters, including a four-week intervention, found that the commute can serve as a transition between home and work roles.⁵ Long morning commutes hurt job satisfaction and increased turnover for people with lower self-control and more work-family conflict, and those who used the trip to think about the work role ahead were less affected.⁵ So the trip has a function. But, as the same paper notes, citing earlier survey work, people rate the morning commute as their least desirable activity of the day.⁵

The time freed up is large. Across 27 countries in 2021 and 2022, working from home saved an average of 72 minutes a day, and workers put about 40 percent of that back into their jobs and about 11 percent into caregiving.⁶ Whatever role-switching the commute provided came bundled with more than an hour of travel. Nothing here shows that a deliberate substitute, a walk before work or a few minutes planning the day, does the same job. It is a reasonable guess, not a finding.

The boundary at the other end of the day may also have moved. Meeting and email data for 3.1 million people in 16 cities during the early lockdowns showed the workday lengthening by 48.5 minutes.⁷ That measured a crisis as much as a work arrangement, so it is a reason for attention rather than a settled result.

What the best test found

If remote work removes this much, it should show up in performance. In the strongest test so far, it did not. At Trip.com, employees were randomly assigned, by even or odd birthdays, either to work from home on Wednesdays and Fridays or to stay in the office full time.⁸ Over the next two years there was no difference in performance grades or promotions, and no difference in the lines of code written by the 653 engineers in the sample.⁸ Quit rates fell by a third, from 7.2 to 4.8 percent, and the reduction was significant for non-managers, women and people with long commutes.⁸ Managers who expected hybrid work to cost about 2.6 percent in productivity ended the experiment estimating a small gain.⁸

This limits the earlier studies more than it contradicts them. The trial measured outcomes, not mechanisms: it shows that two home days a week cost nothing measurable in grades, promotions or code, not that the contact and feedback survived. Three shared days may restore most of the accidental contact. Or the losses that remain may be too spread out, or too slow, to appear in two years of reviews. The study cannot say which.

It is also worth noticing where all of this evidence comes from: Microsoft, a large online retailer's software engineers and a Chinese travel-technology company. Code review leaves a written trail, which is why feedback can be counted there. Whether the same pattern holds for nurses, lawyers or sales teams is not something these studies show.

What remote work took

Put together, the evidence supports a narrower claim than the one this piece started with. The office supplied machinery nobody had to build, chiefly contact across teams and feedback for the asking, plus a reason to stand up. Fully remote work removes much of it, and the cost falls first on people who were learning by being near others, while experienced people get time back. Part-time remote work, on the best evidence, costs nothing measurable in performance; whether it keeps the machinery is still an open question. And the commute, the piece of office life people complain about most, is the piece whose loss is hardest to show matters: what it supplied, a moment to switch roles and some walking, can plausibly be had more cheaply, though no study here tests that.

The practical reading follows from the mechanism. If what matters is unplanned contact, then shared days are worth more when the people who need to learn from each other share the same ones. The retailer study found that when a new hire turned a co-located team into a distributed one, feedback between the original teammates dropped sharply even though they still sat together, apparently because meetings moved online for the one distant colleague.¹ A rule about days in the building is a blunt tool for a problem that is at least as much about who sits near whom.