Last Tuesday I had three meetings: 10:00, 12:30, and 15:00, half an hour each. On paper I gave up ninety minutes of a nine hour day. I finished having written about four hundred usable words and fixed one small bug.
The meetings were not the problem. Ninety minutes is ninety minutes. The problem was the shape they left behind: a 95 minute gap, a 120 minute gap, and a ragged hour at the end, none of which held a single stretch of work that got past the part where I am still remembering what I was doing.
Context switching is expensive because your brain does not hold one task and then simply hold another. It tears down a working configuration and rebuilds a different one, and the rebuild is slower and more error-prone than steady-state work. This is one of the sturdier findings in cognitive psychology: when people alternate between tasks, performance right after each switch is measurably worse than when they stay put, and part of that penalty survives even when they know the switch is coming and have time to prepare. The switch has a price of its own, separate from the time the interruption occupies. You do not just lose the twelve minutes. You lose the twelve minutes, plus the ramp back, plus a quieter tax on the quality of what follows.
Three things happen when you move from task A to task B.
You unload a working set. Whatever you were holding (the shape of the argument, the five variables, the thing you were about to check) is written down nowhere. It is an active pattern, and active patterns do not survive being displaced.
You reload a different one. Task B needs its own context: where you left it, the constraints, what you had already ruled out. That reload takes real time, and more of it the more abstract the task is.
You leave something behind. Part of your attention stays on task A, especially if you left it unfinished, so task B gets a degraded version of you for a while. Sophie Leroy named this attention residue, and the full mechanism is its own article: Attention Residue: Why Your Brain Stays Stuck on the Last Thing. Here the point is narrower. Residue makes switching compound, which is why the fifth switch of a morning feels worse than the first.
Email has almost no working set. Each message is self-contained, so an interruption between messages costs nearly nothing, because nearly nothing was loaded. The same goes for expense reports, scheduling, and most of what we call admin.
Creative and complex work is the opposite. Writing an argument means holding the whole structure at once: what you have claimed, what you still owe the reader, the tone, the thing you are building toward. Hard debugging means holding a model of the system plus a list of hypotheses plus the evidence that killed the last three.
That state is large, fragile, and almost entirely unwritten. A switch dumps it, and when you return you rebuild what you can from your notes. The rebuild is never complete, because the valuable part was not the facts. It was the half-formed shape you were circling, the thing you had not yet articulated well enough to put in the margin.
That is the real creative cost, and it hides well, because you cannot miss what never got made. The day ends, the work is competent instead of good, and you file that under your ceiling. It is often not your ceiling. It is your schedule.
Imposed interruptions arrive on someone else's timing: a colleague at your desk, a call, a child, a meeting. Their defining feature is randomness, so they can land at the worst possible moment.
Self-interruptions are the ones you generate, and their defining feature is not randomness at all. They cluster tightly around friction. Watch yourself for one honest day and you will find you almost never switch away mid-sentence when the writing is flowing. You switch at the exact moment the work turns hard, ambiguous, or boring, which is the moment just before the valuable part.
The two need different fixes. Imposed interruptions are a negotiation and a batching problem: you concentrate them into a window when you are reachable, keep a window when you are not, and produce enough visible output from the second that nobody worries about the silence.
Self-interruptions are not a negotiation, because the other party is you at your least reasonable. Willpower is lowest exactly when the escape is most attractive, which is why in-the-moment resistance has such a poor hit rate. What works is removing the option in advance, while you are clear-headed. A profile configured at 9am is a decision made by a version of you with no stake in the 11:40 moment of difficulty. The friction point still arrives; it finds nowhere to go, and after a few uncomfortable seconds you return to the hard sentence.
Four hours of a day, arranged two ways:
| | Four 55 minute gaps between meetings | One 4 hour block | |---|---|---| | Reload cost paid | 4 times | Once | | Usable minutes after ramp-up | Roughly 140 | Roughly 220 | | Holds a large working set | Barely, and never for long | Yes | | Anticipatory drag | Constant, a meeting is always coming | None |
The middle rows are my rough experience rather than a study, and yours will differ, but the direction is not in doubt. There is also a fourth effect that does not fit in a table: the stretch before a known interruption is nearly worthless. If a meeting starts in eighteen minutes you will not begin anything longer than eighteen minutes, so you do something small, and the pre-meeting gap goes to filler automatically.
A calendar with a meeting every ninety minutes contains, functionally, zero deep work while also holding five free hours. People there conclude they need to work faster. Often they need to move a meeting.
Paul Graham's essay "Maker's Schedule, Manager's Schedule" names this better than anything I have read. Managers run on a schedule cut into one hour units, where any hour can be reassigned. Makers run on units of at least half a day, because that is the smallest unit in which something substantial gets built.
A single meeting in the middle of a maker's afternoon does not cost an hour. It costs the afternoon, by splitting it into two pieces that are each too small to build in. Two consequences worth acting on rather than nodding at:
Standard advice says batch similar tasks. That is nearly right, and the common version is wrong in a way that matters: people batch by topic when they should batch by the mental state the work requires.
Three rough categories. Generative work (writing, designing, architecting, hard debugging) needs a large working set and a long ramp, and is expensive to lose. Evaluative work (reviewing, editing, feedback, planning, triage) needs judgment but little construction, tolerates a short interruption, and runs fine when tired in a way generative work does not. Mechanical work (email, forms, scheduling, small fixes) has a near zero working set and is genuinely fine in fragments.
Sorting this way reshapes a week immediately, because the three have different natural homes: generative in your best contiguous block, evaluative where energy sags after lunch, mechanical in the gaps between meetings. The common failure is exactly inverted. Mechanical work goes first because it is easy to start and clears a small anxiety, which spends the one block that could have held generative work, and then generative work gets attempted at 4pm from a standing start.
Step six is where software earns its place. In Deep Focus I keep a profile called Generative that runs as a whitelist: editor, terminal, one documentation site, nothing else. A blacklist assumes I can enumerate every place my attention might go, and it is more resourceful than that. The Session Planner lays out a day as a sequence of focus and break blocks, so transitions are decided in advance rather than at the moment I feel like moving on, and the Weekly Scheduler activates that profile at a fixed time on fixed days, so the block starts whether or not I feel ready.
Reactive roles. Support, ops, account management, anything on-call. If a real share of your day is generated by other people in real time, a calendar of long protected blocks is fiction, and pretending otherwise just adds guilt. What survives is one defended block at the start or end of the day, with the rest run as an explicit queue.
Jobs where switching is the job. Managers, editors, producers. Fragmentation is the work rather than the failure mode, and the deep block is the exception to schedule rather than the norm to protect. Being interruptible also has real professional value early in a career, so ask for one block rather than a policy.
People who prefer rotation. Some genuinely work better switching between three things, and they are not wrong about themselves. The line worth drawing is between planned rotation, where each task gets a real block, and reactive rotation, which is fragmentation with a nicer name.
There is no reliable single number, and the figures that circulate get quoted with far more confidence than the underlying research supports. The direction is well established: recovery is not instant, it takes longer for complex work, and it takes longer when you left the previous task unfinished. Time your own next few returns and you will have a figure that applies to you.
Because for anything requiring attention you are not doing two things at once, you are alternating quickly and paying a reconfiguration cost on every alternation. The exception is pairing an automatic task with a demanding one, like walking while thinking. Two tasks that both need working memory always run worse together than in sequence, and the quality drop is usually larger than the time drop, which is why it is easy to miss.
Yes, and the reason is the size of the state you reload. Routine work is small self-contained units, so an interruption costs roughly the interruption. Creative work holds a large, mostly unwritten structure in working memory, and a switch discards it, so the same interruption costs the interruption plus the rebuild plus whatever cannot be reconstructed.
Treat it as an environment problem rather than a discipline problem, because self-interruption reliably strikes when the work gets difficult, which is when your resistance is weakest. Decide in advance what is available during a block and remove the rest, so the escape route is closed before you want it. Keep paper next to the keyboard for anything you think of mid-session, since capturing an intrusive thought releases it without paying the switch cost.
None of it felt like a problem while it was happening. Every individual switch was reasonable: the meeting was worth having, the message did need an answer, the tab was opened for a defensible reason. The cost never appears in any single decision. It appears at the end of a week, in the gap between how busy it felt and what is left to show for it.
Get the latest productivity tips and Deep Focus updates delivered to your inbox
