The Office Mandate Is a Hiring Decision: What Five Days a Week Is Actually Costing Engineering Teams

Why Is the Office Question Back on Engineering Agendas?

The office argument was supposed to be settled by now. It is not, and the reason it came back has less to do with collaboration than with three things that arrived at once: expensive leases that nobody wants to write off, a labour market that briefly tilted back toward employers, and a genuinely new argument about AI that deserves to be taken seriously rather than dismissed.

The new argument goes like this. Engineering work has changed shape in eighteen months. A senior engineer now spends a large part of the day directing coding agents, reviewing generated diffs, and deciding what to keep. The craft knowledge that makes that work good, when to trust the output, when the abstraction is wrong, when a passing test suite is lying to you, is tacit and hard to write down. Tacit knowledge transfers best in person. Therefore, the argument concludes, get everyone back in the building while the discipline is being invented.

It is a coherent argument and it is worth engaging with rather than mocking. The problem is that it is being used to justify a policy whose measured effects run in the opposite direction, and the effects fall hardest on precisely the people the argument is about. A mandate does not select for engineers who transfer tacit knowledge well. It selects for engineers who cannot leave.

That is the frame this piece takes. Not whether offices are good, which is unanswerable in the abstract, but what a five-day mandate does to the specific population of people you need in an engineering organisation in 2026, and what the numbers say happens next. As the MIT Sloan Management Review analysis of mandates and high performers notes, office utilisation across the past year has hovered around 50% of pre-pandemic norms even as policies tightened, which tells you something about how much of this is policy on paper rather than behaviour in practice.

What Does the Evidence Say About Who Actually Leaves?

The most useful study on this question did not ask anybody how they felt. It watched what they did. A working paper by researchers at the University of Pittsburgh, Return to Office Mandates, Brain Drain and Gender Difference, reconstructed the employment histories of more than three million workers from LinkedIn profiles and tracked what happened at S&P 500 firms before and after those firms imposed mandates. It reports a 14% increase in employee turnover following a mandate, concentrated among senior, highly skilled, and female employees.

The hiring side is the part engineering leaders tend to miss, and it is arguably worse than the attrition. The same study, summarised in HR Dive's coverage of the brain drain findings, found that the time taken to fill a vacancy rose by roughly 23% after a mandate, while the overall hire rate fell by 17%. So you lose more senior people, and you replace them more slowly, and you replace fewer of them. Those three effects compound rather than offset, and they compound into your delivery roadmap two quarters later where nobody attributes them to the policy.

The survey evidence points the same way with a sharper edge on seniority. Gartner's survey of 2,080 knowledge workers measured intent to stay under strict on-site requirements and found it 8% lower among employees generally, 16% lower among high performers, 11% lower among women and 10% lower among millennials. High performers respond at double the rate of the average employee. That is not a rounding difference; it is the entire retention problem stated in one number.

Set against this, the strongest evidence for the alternative is not an opinion survey either. The largest randomised controlled trial of hybrid work, published in Nature as hybrid working from home improves retention without damaging performance, randomly assigned 1,612 graduate employees in engineering, marketing and finance roles at a large travel technology company to either five days in the office or three days in and two at home. Attrition in the hybrid group fell by roughly a third. Performance reviews, promotion rates and productivity showed no measurable difference across two subsequent years of review data. A third less attrition, for nothing.

None of this proves an office has no value. What it establishes is a price, and it is a price most engineering organisations have never actually written down before signing the policy.

Why Does a Mandate Hit Senior Engineers Hardest?

The concentration of attrition among senior people is not a coincidence, and it is not primarily about preferences. It is a selection effect with a simple mechanism: the cost of leaving is lowest for the people with the most options, and in engineering, options track seniority almost perfectly.

A staff engineer with twelve years of experience, a public track record and three recruiters in their inbox each week faces a very different decision from a two-year developer who has not yet built that optionality. When the policy changes, both are unhappy. Only one of them has a credible exit within a fortnight. A mandate is therefore an unusually precise instrument for removing exactly the people you cannot afford to lose, while retaining exactly the people who most need the mentoring those leavers were providing.

There is a second mechanism that is specific to senior engineering work and less often discussed: life stage. Fifteen years into a career, people have children in a particular school, a partner with a job of their own, a mortgage in a particular town, and often a caring responsibility for an ageing parent. The flexibility is not a lifestyle preference for this group. It is the load-bearing structure that makes a demanding job compatible with the rest of their obligations. Remove it and they do not negotiate, they relocate their employment.

This matters more in 2026 than it did in 2021 because of what senior engineers now do. When a large share of your code arrives from an agent, the constraint on your organisation shifts from typing capacity to judgement capacity, the ability to look at a plausible-looking two-hundred-line diff and know that the error handling is wrong for your domain. That judgement lives almost entirely in your senior population. An organisational policy that disproportionately sheds seniors is, in the current environment, a policy that sheds review capacity at the exact moment review capacity became the bottleneck.

The uncomfortable version of this is that a mandate can look successful for two quarters. Headcount falls without a redundancy programme, costs come down, and nobody has to stand up in a town hall. The bill arrives later, in incident frequency, in review latency, in the quiet degradation of code nobody senior looked at closely.

What Happens to Your Hiring Pool the Day You Post a Location?

Attrition is the visible half. The hiring pool is the invisible half, and it is the half that determines what you can build over the next three years.

The moment a role requires five days in a specific building, the addressable candidate pool collapses to people already within commuting distance of that building, plus the small number willing to relocate for it. For a specialist role, that is often catastrophic. If you need someone who has run Kafka at scale, has shipped a regulated fintech product, and can operate an evaluation harness for LLM features, there may be a few hundred such people in Europe and a handful within forty minutes of your office. The mandate did not make hiring harder. It made hiring a different, much smaller problem.

The developer population itself has not gone back. Stack Overflow's 2025 developer survey work section puts 32.4% of developers fully remote against 17.9% fully in-person, with the remainder hybrid or choosing for themselves. Fully in-person is the smallest category by a wide margin. Country-level breakdowns in the same data show 45% fully remote in the United States and 22.5% in Germany, so the shape varies, but nowhere does the in-office group approach a majority.

There is a compounding effect that shows up in offer acceptance rather than in application volume, and it is easy to misread. Your pipeline may still look full, because plenty of people apply to plenty of things. What changes is the mix. The candidates who accept a five-day on-site offer are, on average, the ones with fewer competing offers. You are not sampling from the same distribution any more, and no amount of interview rigour recovers a candidate who never applied.

For European engineering organisations, the geography of this is unusually stark. Hybrid arrangements are close to standard across most of Western Europe, and a strict mandate stands out as an outlier rather than as a norm being restored. When your policy is more restrictive than the market you are recruiting into, the policy is a differentiator, just not in the direction the slide deck claimed.

Does Colocation Actually Make AI-Era Engineering Better?

This is the question the honest version of the debate turns on, and it deserves better than either side usually gives it. There is something real behind the colocation argument. Tacit knowledge does transfer through proximity. Ambient overhearing does catch mistakes. A whiteboard conversation does resolve in twenty minutes what a comment thread stretches across three days. Anyone who has worked both ways knows this.

The mistake is treating the office as the only delivery mechanism for those benefits, and then buying the whole package to get them. Almost every one of those benefits comes from a smaller, cheaper, more targeted ingredient: shared working hours, high-bandwidth synchronous time when it matters, and enough shared context for a quick conversation to be productive. Those ingredients can be bought without a five-day mandate, and the randomised evidence linked above suggests three days in and two at home retains most of them while removing the retention cost.

What is genuinely new is that agentic engineering pushes in the opposite direction to colocation, and this is the part the pro-office argument has not caught up with. When agents write a large share of the code, the work that decides quality is specification, review and verification. Specifications are written artefacts. Reviews are asynchronous by nature and better for it, because a careful reviewer needs uninterrupted time far more than they need proximity. The teams that are getting the most out of coding agents are the ones that write things down: clear tickets, explicit acceptance criteria, architectural decision records, runbooks. That discipline is exactly what distributed teams are forced to build and colocated teams are permitted to skip.

There is a deeper irony here. A colocated team can survive on hallway context because a human can always ask the person who knows. An agent cannot walk down the hall. It reads the repository, the docs and the specs you wrote, and its output quality is bounded by them. Organisations that never had to write anything down are discovering that their institutional knowledge was stored in a form their new tooling cannot access. Distributed teams wrote it down years ago because they had no choice.

So the AI argument for the office is not wrong about the value of tacit knowledge. It is wrong about the direction of travel. The systems now consuming your engineering knowledge are text-based, and the discipline that makes them effective is the discipline distribution imposes.

Is the Mandate a Talent Strategy or a Severance Strategy?

There is a version of the office mandate that is not about collaboration at all, and pretending otherwise makes the rest of the conversation dishonest. A mandate that produces voluntary departures reduces headcount without a redundancy consultation, without severance, and without a press release. Where that is the intent, the attrition figures are not a cost of the policy. They are the policy.

Engineering leaders should be clear-eyed about this for one practical reason: it does not stay secret, and the reputational cost lands on recruiting rather than on the executives who chose it. Developers talk. Salary and policy information circulates through Slack communities, Blind, Reddit and former colleagues faster than any employer brand campaign can respond to. A company understood in its local market as having used attendance policy to shed staff will pay for that understanding in every subsequent senior offer it makes, often for years.

The second practical problem is that attrition-by-policy is untargeted. A redundancy process, whatever else can be said about it, lets you decide who leaves. A mandate lets the market decide, and the market takes your most employable people first. If you actually need to reduce engineering headcount, doing it deliberately preserves the capability you meant to keep. Doing it through the car park does not.

None of this argues against reducing cost. It argues against reducing it through a mechanism that selects your leavers by their outside options rather than by your strategy, while telling everyone it is about collaboration.

What Actually Produces the Benefits People Attribute to the Office?

If you strip the argument back to mechanisms, the office is a bundle of four things: shared working hours, occasional high-bandwidth contact, shared context, and visible accountability. Each can be provisioned separately, and provisioning them separately is what strong distributed engineering organisations do deliberately rather than hope for.

Shared hours are the one that is genuinely non-negotiable, and it is the one most often confused with location. A team spread across a four-hour band with six hours of overlap behaves nothing like a team spread across a twelve-hour band with ninety minutes of overlap. In the first case a question asked at eleven gets answered before lunch and a production incident gets two engineers on a call immediately. In the second, every exchange costs a day, and the cost is invisible in the contract and enormous in the delivery plan. This is the single most important variable in distributed engineering and it has nothing to do with whether anyone is in a building.

High-bandwidth contact is real and periodic rather than continuous. Quarterly gatherings of a few days, structured around planning, architecture and the unstructured social time that makes later remote conversation easier, deliver most of the relationship benefit at a fraction of the cost and disruption of permanent colocation. The failure mode is not distribution, it is distribution with no in-person contact at all, ever.

Shared context is a written artefact problem and it responds to investment. Decision records that explain why, not just what. Onboarding documentation that someone actually followed last month. Tickets with acceptance criteria a stranger could implement against. A repository whose structure and conventions are legible without a guided tour. This is the same work that makes coding agents effective, which means it now pays for itself twice.

Visible accountability is the one most often used as a euphemism for surveillance, and it is the easiest to get right. Shipped software is visible. Pull requests are visible. Incident response is visible. Any engineering organisation that cannot tell who is contributing without watching them at a desk has an observability problem in its own delivery process, not a location problem, and installing badge readers will not fix it.

How Do Nearshore Teams Fit This Picture?

Once you accept that overlap rather than location is the binding constraint, the practical question changes. It stops being where do we make people sit and becomes where can we find senior engineers who will work our hours, under our legal regime, and stay long enough to matter.

That is the question nearshore answers well, and it is worth being precise about why. A dedicated team in Serbia working with a client in Germany, the Netherlands, the UK or the Nordics shares most of the working day. Central European Time against UK time is one hour; against the Nordics and most of Western Europe it is zero. That is not a partial overlap to be managed with handover documents, it is the same working day. Compare that with an offshore arrangement where the overlap is ninety minutes and every clarification costs a full cycle, and the difference shows up in cycle time long before it shows up in anyone's satisfaction survey.

The talent argument is the one that connects directly to the mandate problem. Western European engineering organisations imposing strict on-site policies are shrinking their pool at the same moment they need more senior judgement than ever. Central and Eastern Europe has a deep, replenishing pool of engineers with strong mathematical and systems foundations, and in Serbia specifically a sector large enough to staff genuinely senior teams rather than a thin layer of leads over a mass of juniors. Hiring there does not require anyone to relocate, uproot a family, or trade their working life for a commute.

For European clients there is a legal dimension that matters more each year. Working inside the EU regulatory perimeter means GDPR obligations, data residency, the NIS2 and Cyber Resilience Act regimes, and AI Act duties are shared constraints rather than contractual translations across a jurisdictional gap. When your partner is subject to the same rules you are, the compliance conversation starts from the same defaults instead of from a data processing addendum nobody wants to test in court.

This is how we build teams at Stepto. Dedicated engineers in Serbia, working the client's hours, embedded in the client's rituals and repositories, staying with a product long enough to develop the tacit knowledge the office argument is really about. Not a ticket queue in a distant time zone, and not a body shop that rotates people out every six months, because both of those destroy exactly the continuity that makes senior engineering valuable. The reason we care about long engagements is the same reason the attrition research matters: an engineer who has owned a system for two years makes better decisions about it than one who has read the documentation, and no policy about seating arrangements substitutes for that.

The honest framing for a CTO weighing this is not nearshore versus in-house. It is a capacity question. If your mandate is going to cost you senior people and lengthen your time-to-fill by roughly a quarter, you need a source of senior capacity that does not depend on your commute radius. A dedicated nearshore team is one credible answer, and it happens to be one that gets stronger the more written, reviewable and specification-driven your engineering practice becomes.