← Blog

Focus, explained

How to take a break without losing the thread

Almost every argument about breaks is an argument about length. Twenty-five minutes or fifty. Five off or fifteen. It is the wrong end of the problem. The research that has anything concrete to say is not about the break at all — it is about the return, and it suggests the cost of coming back is largely decided in the couple of seconds before you leave.

By ComunAI · 7 September 2026 · ~6 min read

First: breaks are not the enemy of focus

Worth settling before anything else, because the deep-work literature is sometimes read as an argument for never stopping. Atsunori Ariga and Alejandro Lleras had people perform a long, monotonous vigilance task and found that brief, occasional switches to a different activity prevented the usual decline in performance over the session.¹ Evidence: Moderate We covered the vigilance side of this in how long a deep work session should be. Their interpretation was that the goal itself deactivates with prolonged use, and a brief departure lets it be reactivated.

At the aggregate level, Patricia Albulescu and colleagues meta-analysed the micro-break literature and found a reliable benefit for wellbeing and vigour, with a smaller and more variable effect on performance that depended on the kind of task.² Evidence: Moderate Note the asymmetry, because it is the part usually skipped: the evidence that short breaks make you feel better is stronger than the evidence that they make you produce more.

The part almost nobody costs: getting back in

Erik Altmann and Greg Trafton built a model of how we hold a task goal, and its central claim is uncomfortable. A goal is not stored somewhere and retrieved intact; it is an activation level that decays, and resuming means retrieving it in competition with everything else that has since become active.³ Evidence: Moderate

That predicts something specific and testable: the return should cost time proportional to how far the goal has decayed and how much interference has arrived. Altmann and Trafton measured the timecourse directly and found recovery from an interruption is not instantaneous — the first actions after resuming are slower and more error-prone, and performance climbs back over a period rather than snapping back.⁴ Evidence: Moderate

Gloria Mark, Daniela Gudith and Ulrich Klocke found the same shape in real offices, with an additional twist: interrupted work was often completed faster, but at the cost of higher stress and effort.⁵ Evidence: Moderate People compensate. The compensation is not free, and it is invisible in output measures.

The cheapest intervention in this whole literature

Here is the finding that changed how we think about the timer. Greg Trafton and colleagues gave people a warning before an interruption — a short window before the task was taken away — and found that resumption was faster afterwards.⁶ Evidence: Moderate The interval was used to encode where they were: what they had been doing, what came next.

This follows directly from the memory-for-goals account. If the return is a retrieval problem, then anything that leaves a retrieval cue behind makes the return cheaper. A few seconds spent stating your place is a few seconds spent building the cue you will need.

The practical form is unglamorous and it is roughly the only concrete technique this research supports:

• Before you stand up, write the next action. Not the task. The next physical move — "finish the second paragraph of the summary", not "work on the report". A goal stated at that resolution is the retrieval cue; a topic is not.
• Leave the work mid-thought if you can. Stopping at a clean boundary feels tidier and gives you nothing to retrieve. Stopping mid-sentence leaves the context sitting there, visible.
• Take the break at a moment you chose. A break you decided on comes with a warning interval by definition. An interruption does not, which is most of why it costs more.

What to do during it — the shorter, weaker section

This is where the evidence thins out considerably, and where most advice becomes confident in inverse proportion.

Sabine Sonnentag and Charlotte Fritz developed the standard instrument for recovery experiences and identified psychological detachment — mentally disengaging from work — as the component most consistently associated with recovery.⁷ Evidence: Moderate This is correlational, survey-based occupational research, and it is mostly about evenings and weekends rather than the ten minutes between two blocks of work.

Emily Hunter and Cindy Wu surveyed workday break activities and reported that breaks taken earlier in the day, and breaks doing something the person preferred, were associated with better outcomes — while finding no benefit for longer breaks as such.⁸ Evidence: Early A single field study using self-reports. Treat the direction as a hypothesis, not a prescription.

Marc Berman, John Jonides and Stephen Kaplan found that a walk in a natural setting improved subsequent performance on an attention-demanding task more than a walk in an urban one.⁹ Evidence: Early A real and much-cited result, on small samples, and the broader attention-restoration literature has had a mixed replication record since. We are not going to tell you a walk in the park fixes your afternoon.

The one thing this section does support: if the break involves picking up a second stream of unfinished obligations — a work chat, an inbox, a feed of things that want something from you — you have not detached, you have switched tasks, and you now have two open threads instead of one. That is the reliably bad option in a field of weak ones.

Where the evidence stands

The calibrated bottom line

Take breaks. The evidence that they help how you feel is decent, the evidence that they protect performance on long monotonous work is reasonable, and the evidence for any particular schedule is absent.

Then spend your effort on the return rather than the departure. The resumption cost is real, it is measurable, and it is the one part of this that a small deliberate habit demonstrably reduces. Say where you are before you go. In writing, in one line, at the resolution of a next action.

And notice what that implies about the break you did not choose. Every result in the resumption literature gets worse when the departure is involuntary, because an interruption is exactly a break with the preparation interval removed. Which is the argument for controlling what can interrupt you in the first place — the subject of the post on app blocking and the one on notification batching.