How to remember what you were doing after an interruption

You get pulled away for five minutes, come back, and the thing you were doing is gone. The interruption was short; losing your place is what costs the time. Here is why it happens, what research on interruptions has found, and a habit that makes getting back much easier.

The short answer

Before you switch away, write down one line about where you were and what you were going to do next. When you come back, read it before doing anything else. Keep one list of everything you've started and not finished, so the task you were pulled away from can't disappear from view.

Why you lose your place

While you work on something, you hold its current state in working memory: which file, which step, what you were about to try. Working memory is small and fades quickly. An interruption fills it with something else, and when you return there's nothing on screen to tell you where you stopped, so you have to rebuild it. For people with ADHD, whose working memory is under more strain to begin with, this happens more often and costs more.

Context switches don't have to come from outside. Following a tangent, checking a notification or starting a second task while the first is waiting all leave the same kind of gap. Running several AI agents at once adds more: each one finishes at its own time and asks for your attention.

What the research says

Part of your attention stays behind

Sophie Leroy named this attention residue: when you switch from an unfinished task, part of your attention stays on it, and you do worse on the next task as a result (Leroy, 2009).

A plan for coming back reduces it

In a later study, people who took a moment before switching to write down where they were and what they would do next, which the authors call a ready-to-resume plan, showed less attention residue and did better on the interrupting task (Leroy & Glomb, 2018).

Cues left behind help you resume

Experiments on task interruption found that people resume faster when a cue from just before the interruption is still available, and that people spontaneously try to prepare such cues when they see an interruption coming (Trafton, Altmann, Brock & Mintz, 2003).

Working faster isn't free

In a study of interrupted office work, people finished interrupted tasks faster and with no loss in quality, but reported more stress, frustration, time pressure and effort (Mark, Gudith & Klocke, 2008). Compensating works, but you pay for it.

A five-step habit for getting back on track

1. Name what you start. When you begin anything you'd recognize by name, write a few words for it. Starting is when you have the most context and the least reason to think you'll forget.
2. Leave a note when you switch. One line: where you stopped, and what you were about to do. This is the ready-to-resume plan from the research above. Fifteen seconds is enough.
3. Park tangents under their parent. When a task spawns a side task, write the side task down under the original, so finishing the tangent leads you back instead of somewhere new.
4. Look before you pick something new. When you come back, read the list of unfinished work first. Start with whatever has waited longest.
5. Close things when they're done. Remove finished items right away, so the list only shows what is still open and stays short enough to read at a glance.

Tools that make the habit stick

A text file or a sticky note can do all five steps. The difficulty is that each step has to be quicker than the interruption, or you'll skip it, and a text file doesn't tell you which item has been sitting there all afternoon.

Mindthreads is a free Mac app built for this habit. ⌃⌥N captures a thread from any app in about two seconds. c adds a tangent as a child of the selected thread. Threads warm in color the longer they go untouched and pulse once they've been neglected, so the task you were pulled away from stands out when you return. x completes a thread and clears it off the board.

If your interruptions are mostly your own tangents, the ADHD page goes into how the app handles those. If they come from AI agents finishing work, the MCP server lets the agents record their own threads.