Where your engineers work matters less than whether they chose it

Every few months, another headline: Amazon mandates five days in office, productivity will improve. Stanford publishes data showing hybrid workers have 33% lower attrition with no measurable drop in output. A Gallup survey finds 51% of remote-capable employees are hybrid and intend to stay that way.

So which is it? And what should an engineering leader actually do with contradictory data?

The confusion is not a measurement problem. It is a question-selection problem. Most of the debate treats location as the key variable (remote, hybrid, in-office) when the research increasingly points to a different one: who gets to decide.

The large-sample data is more consistent than the headlines suggest

Nick Bloom at Stanford has produced the most rigorous independent dataset on this question. In a preregistered experiment with more than 1,600 employees at Trip.com, workers assigned to two days of home working per week showed no measurable difference in performance ratings, no difference in promotion rates, and 33% fewer resignations compared to full-office peers. The study, published in Nature in 2024, has the methodological properties — randomised assignment, pre-registration, and a follow-up period long enough to detect attrition effects — that most HR surveys lack.

The WFH Research tracker, which Bloom co-founded, shows remote work stabilised at around 25% of workdays in 2025 and has held there. It did not collapse under RTO mandates, and it did not expand. Gallup's Q2 2025 data tells the same story: 51% of remote-capable employees are now hybrid, a share that has been roughly stable for two years.

What the large-sample studies consistently show across different countries, industries, and methodologies:

• Hybrid, typically two to three days remote per week, performs as well as full-office on most measured productivity indicators.
• Hybrid outperforms full-office on retention by a significant margin. The 33% figure from Trip.com is within the range of broader surveys.
• Full-remote shows mixed results depending on task type: experienced engineers on primarily asynchronous work perform comparably to hybrid; roles with intensive synchronous collaboration requirements show more degradation over time.

The variable that actually explains the variance

This is where the research stops matching the headlines.

Most coverage treats hybrid as a single condition. It is not. There are two meaningfully different versions. The first is preference-based hybrid: the employee decides which days, how often, and from what baseline. The second is mandated hybrid: the employer specifies the schedule — you are required in Tuesday through Thursday.

These produce different results. Research measuring employee satisfaction on a 10-point scale found preference-based hybrid averaged 8 out of 10; mandated hybrid schedules scored significantly lower. The mechanism is straightforward. When engineers have autonomy over their arrangement, they use it according to their actual work patterns — deep-work days at home, in-office days for sprint planning and design reviews. When the same arrangement is imposed, the synchrony benefit shrinks (people who do not need to collaborate are compelled in on the same days) while the cost burden stays the same.

71% of hybrid workers report feeling more in control of their working lives, compared to 35% of full-office workers. Perceived control and autonomy are consistently predictive of lower burnout, lower attrition, and higher sustained output in knowledge-work research. These are not feel-good metrics. They are leading indicators of the attrition figures that arrive six to eighteen months later.

What RTO mandates actually measure

Companies that have issued return-to-office mandates have generated some of the most useful natural experiments on this question. The aggregate picture: firms announcing RTO mandates see an average 13 to 14% increase in abnormal turnover in the months following. 80% of companies with active RTO policies report losing talent as a result. Nearly half reported higher attrition than anticipated.

The pattern of who leaves is the important part. It skews senior. 36% of senior-level job seekers cited RTO mandates as a reason for leaving their last role. Baylor University research confirmed this: mid- and top-level managers exit at higher rates than junior staff following RTO announcements. The people with the most outside options (senior engineers, staff engineers, tech leads) are the ones most likely to take them.

A BambooHR survey of more than 1,500 US managers found that a quarter of C-suite executives hoped for some voluntary attrition after implementing RTO. One in five HR professionals admitted their in-office policy was partly designed to make some staff choose to leave. This explains why some organisations were comfortable with the initial churn but surprised when it continued longer or skewed more senior than expected.

The productivity rationale for RTO mandates is weaker than most announcements imply. When researchers examine the studies cited in support of full in-office attendance, they typically find correlational evidence (companies performing well also happen to have called workers back), rather than causal evidence. The controlled experiments and natural experiments with clean comparison groups consistently fail to find a productivity advantage from mandating full in-office attendance for knowledge-work roles.

Engineering teams feel this more than most

Domain specifics matter here. Engineering work has a higher proportion of cognitively intensive, low-interruption tasks than most office functions. Writing a database migration, debugging a race condition, reviewing an architectural diagram — these benefit from uninterrupted blocks of two to four hours in a way that sales coordination or client relationship management typically does not.

What engineers specifically need from in-person time is narrow: high-bandwidth design discussions, onboarding where informal explanation matters, hard architectural decisions where whiteboard dynamics help, and the social trust-building that makes async collaboration function over time. A three-day-a-week in-office mandate does not add more of this. It adds meetings, background noise, and the expectation of availability during hours that might otherwise have been used for focused work.

ADP Research Institute data for 2025 shows tech roles already have the highest remote and hybrid adoption of any sector — 29% hybrid, 13% fully remote. When in-office mandates arrive in engineering organisations, the people most likely to leave are those who structured their lives around the flexibility and those with enough skill to find another role that preserves it. Neither group is easy to replace.

A secondary cost is hiring. A restrictive in-office requirement narrows the candidate pool for every open role. In a market where experienced backend engineers and staff-level engineers are in consistent demand, this shows up as longer time-to-hire and lower offer acceptance rates. The full cost of an RTO policy includes not just who leaves but who you fail to attract.

What to actually decide

The question worth asking: What do you need in-person time to accomplish that you cannot do asynchronously? For most engineering teams, the honest answer is sprint kickoffs and retrospectives, quarterly planning sessions, onboarding sessions for new engineers, and the occasional architectural decision that benefits from whiteboarding. That list might total 20 to 40 days per year, not 200.

The structure that tends to hold up: Team-level autonomy with shared anchor days. Each team agrees on two to four days per month when in-person attendance is expected. Those days are reserved for the work that in-person actually improves. The remaining days are the individual's to allocate. This captures the coordination benefit without a rigid daily schedule that ignores varying working patterns within the team.

What to measure: If you change your policy, track attrition rates by arrangement type, time-to-hire for open roles, and self-reported deep-work hours. Attrition by arrangement tells you whether the policy is driving the people you want to keep toward the door. Time-to-hire tells you whether location requirements are narrowing the candidate pool. Direct productivity measurement is hard to standardise across engineering roles; these proxies are more tractable and more honest about what the policy is actually affecting.

The data from the last five years has largely settled the headline question. Hybrid works. Full-remote works for the right roles and the right teams. Full-office mandates for senior knowledge workers generate attrition without a corresponding productivity gain. The remaining interesting question is narrower: what kind of hybrid, designed by whom, producing what specific coordination outcomes? That question is different for every team. It is worth asking carefully before the people who could have helped answer it have moved to companies that did not need to ask.