An interrupted task wants to finish itself

In the late 1920s, a young researcher named Maria Ovsiankina ran a deceptively plain experiment. She gave people simple tasks — assembling a shape, working a small puzzle — and then interrupted them partway through, often by manufacturing some excuse to take the materials away. Then she waited to see what they’d do. Not what they’d remember. What they’d do.

What they did was pick the task back up. Without being asked. Without a reward for finishing. Given a free moment and the chance, most people returned to the thing they’d been pulled away from and completed it, as if it had a small gravity of its own.

Her mentor, Kurt Lewin, had a way of describing this. Starting a task, he argued, sets up a kind of tension in you, he called it a quasi-need, and that tension doesn’t discharge until the task is done. Interrupt the task and the tension just… stays. It sits there, quietly loaded, until something lets it resolve. The pull back to the puzzle is that tension looking for its exit.

This is a close cousin of the more famous Zeigarnik effect, and builders constantly blur the two. Zeigarnik is about memory: we recall interrupted tasks better than finished ones. Ovsiankina is about motion: we feel a drive to resume them. That difference matters when you’re designing, because you don’t just want your users to remember the thing they didn’t finish, you want them to come back and finish it. Zeigarnik keeps the loop lit in memory; Ovsiankina is the force that walks them back to it.

Here’s the build decision hiding in that. Most products treat an interrupted flow as a failure — the user bailed on signup, abandoned the cart, left the profile half-built — and quietly delete the evidence. A fresh visit gives them a blank slate. But a blank slate throws away the tension. You’ve swept away the exact thing that would have pulled them back.

The alternative is to make the unfinished state visible and resumable. The half-written draft that’s still there tomorrow. The “pick up where you left off” that drops them one step from done, not back at the start. The profile that says 60% complete with the three remaining fields already waiting. None of these are new mechanics, they’re all just refusals to erase an interrupted task. You’re not manufacturing motivation. You’re declining to destroy motivation the user already generated by starting.

There’s an ethical seam here worth naming, because this is a real force and real forces get abused. The honest use of the resumption drive is to help people finish things they chose to start — the report, the setup, the thing that pays them back when it’s done. The dishonest use is to open loops the user never wanted, purely so the itch drags them back to your surface. A saved draft of their own work is a gift. A fake “you have 1 unfinished step” on a task that benefits only your engagement graph is a splinter you jammed under their fingernail. The mechanism doesn’t know the difference. You have to.

The quiet lesson from a puzzle taken away in 1928 is that unfinished has energy. Your users hand you that energy every time they start something and get pulled away — a meeting, a notification, a kid, a train. The only question is whether your product keeps the unfinished thing alive so they can spend that energy finishing, or wipes it clean and makes them summon it all over again from zero.

The science, to look up: the Ovsiankina effect (Maria Ovsiankina, 1928), Kurt Lewin’s tension-system account of quasi-needs, and its contrast with the Zeigarnik effect (Bluma Zeigarnik, 1927), memory of the open loop vs. the drive to close it.