The night shift nobody works
A team spread around the world can keep one task moving all day and all night. It buys the days with hours, and the places where it lasts are the ones that wanted to stop working at night.
· 4 min read
Put three engineers of equal standing eight hours apart, in California, Ireland and Singapore, and a hard problem need never sit still. Each works a normal day and passes the work west at dusk. Nobody holds the work until headquarters wakes up, because there is no headquarters. The practice is called "follow the sun", and in theory two sites cut a project's duration by half and three sites cut it by two thirds.1
The clock in that arithmetic is right. The bill is missing. Every pass of the work is a handoff, and each handoff adds a person to the task. People per task is also how distance slows software down, in the best measurement of it. A 24-hour team is faster in days and more expensive in hours, and it pays only where days are scarcer than hours.
What distance costs
At Lucent, James Herbsleb and Audris Mockus compared changes made inside one site with changes that crossed sites, in one department spread across the UK, Germany and India.2 A change that crossed sites took about two and a half times as long to finish: in one department about 5 days against 12.7, in another about 7 days against 18.2 Distance was not the direct cause in their model.2 Changes that crossed sites involved more people than comparable same-site changes, and the number of people involved was strongly related to how long a change took.2
In their survey, engineers reported being held up for information about as often by distant colleagues as by local ones, but each delay that crossed sites lasted almost a day and a half longer.2 Over two and a half to three and a half years of data, neither department reduced how much its sites depended on one another.2
Those sites shared part of their working day. Follow the sun removes the overlap on purpose, and puts three people on every task that runs around the clock. When a question comes up after a handoff, the person who knows the answer went to bed at the last pass and is back in sixteen hours.
GitLab's handbook describes what this costs when it is done with code.3 In "Follow-the-Sun Pairing", people in EMEA, APAC and the Americas take turns on one deliverable, which the handbook calls "less efficient but faster" than one person doing the work.3 The handoff must be context-rich: the first pair describes every design decision in detail, so that the next pair knows which choices were deliberate and does not redo them.3 The handbook contrasts this with code review, where a reviewer who lacks context can simply ask a question.3
That contrast is the mechanism. A question is cheap when the person who can answer it is awake. When the answer is a night away, the safe move is to write everything down before leaving, and a handoff costs as much as writing out a working mind. GitLab names a coordinator for these handoffs and recommends the practice for escalations and high-priority work, such as critical customer fixes, high-severity security issues and corrective work after serious incidents, with the urgency confirmed by leadership such as the engineering manager and product manager.3
The trade is real, and it has a price. In the one quasi-experiment reported, teams working this way finished faster than collocated teams, while the time spent on development rose by about 10%.1 The literature describes the practice as fit for a product needed quickly "and the cost is irrelevant to the client".1 Few companies have kept paying. Many tried follow the sun and abandoned it.1 IBM's first five-site attempt struggled with its daily handoffs and dropped follow the sun while keeping the distributed team.1 Three later IBM projects split between the United States and India, on one code base, were dropped.1 Managers at Infosys said they had never run a project this way, and experts at Intel considered it unfeasible for software development.1
One IBM project with sites in the United States and Australia succeeded, after the two distant teams were merged into one cohesive team.1 That is the nearest thing in the record to a team of equals. The Lucent data show what goes missing without it: distant colleagues reported less sense of being one team, and judged that people at other sites were less likely to help when workloads were heavy.2 Equal standing makes the pass possible, because people take over work from colleagues they trust. It does not make the pass cheap.
Where it lasts
When a service has enough work to grow its team, Google's site reliability engineers prefer to split the on-call team between sites, and one stated reason is that a follow-the-sun rotation lets the team avoid night shifts altogether, which harm health.4 The same book says such teams pay communication and coordination overhead.4 What passes between shifts is small: an incident and its follow-up take about six hours on average.4
Overnight radiology is the older case. Nighthawk Radiology was founded after William Bradley, lecturing in China, read an emergency case for his California office and realized he was reading it with a fresh perspective in daylight, while in California it was early morning.5 In 2000, night coverage was costing each partner in his practice $30,000 a year.5 By 2004 the company had offices in Sydney and Zurich and served 500 hospitals, with preliminary reads that the requesting practice followed with formal ones.5
In both cases a whole piece of work is passed, a scan or a shift of incidents, and handed back finished. What these services sell is sleep: the night still gets worked, by someone for whom it is day.
For ordinary development the trade usually runs the other way. One engineer who keeps the whole problem in one head, and sleeps on it, may finish later and spend fewer hours getting there. No one has measured the same project done both ways over months in a real company, and the firms that tried and stopped had little reason to publish.
A company that runs follow the sun on everything is declaring every task an emergency, and paying the emergency rate for each.
Sources
- 1
- 2
- 3
- 4
- 5
About this piece
- Kind
- Research essay. An argument grounded in published research.
- Evidence
- Moderate
- Research
- Standard, 5 sources
- Made
- My idea. AI researched and wrote it, and an AI model checked each sourced claim against its source before I published it here.
- Updated
- Nov 21, 2026