What It Means for a Tool to Respect Your Time

In 1996, researchers at Xerox PARC hung an eight-foot piece of string from the ceiling. One end connected to a small motor; the other end just hung there. The motor was wired to a nearby Ethernet cable, so when data flowed through the network, the string twitched. When the network was quiet, so was the string.

This was not a prototype for a product. It was a demonstration of an idea. Mark Weiser and John Seely Brown called it the Dangling String, and it was their example of calm technology: a system that informs without demanding attention. The string lived at the edge of your awareness. You didn't have to look at it. But if you glanced over and saw it moving fast, you knew something was happening. You could choose whether to care.

Nobody builds the Dangling String anymore. They build notification banners, red badges, animated tooltips, and modal dialogs that require a decision before you can continue. Software that constantly checks in to ask: are you still there? Did you see this? Do you want to do this now?

What "helpful" usually means

There is a theory behind intrusive software. It goes like this: if something is important, you want to know about it immediately. The tool that tells you the most, fastest, is the most helpful tool.

This theory fails in practice. It takes an average of 25 minutes to refocus after an interruption — Gloria Mark's finding from years of research at UC Irvine, cited often because it lands with appropriate weight. The ping asking you to update, the banner asking if you meant to close that tab, the assistant bubbling up to suggest something: each costs more than the interruption itself. You context-switch, you evaluate, you make a decision. Then you try to find your place again.

The problem isn't that software wants to be helpful. The problem is that it treats every moment as equally valid for interruption.

Peripheral attention

Weiser and Brown's insight was that human attention has two modes: the center, where you focus, and the periphery, where you passively register things without effort. A skilled tool occupies the periphery. It does its work quietly and appears at the center only when you need it — then retreats.

This is how a good colleague works. They don't interrupt you every time they complete a step. They do the work, note what they found, and surface it when there's something worth your attention. The question "would you like me to update the project notes from our call?" arrives at the end of the meeting, not during it.

Most software hasn't learned this. But a few patterns have.

Email clients prefetch your inbox overnight, so it's ready when you open the app. Most music players cache the next track before the current one ends. These are small acts of respect. The software did its work while you weren't watching and surfaced the result only when it was relevant. When Google Maps detects you've missed a turn, it recalculates in the background and updates your route quietly. It doesn't send a message announcing it's thinking.

The opposite is a progress bar that requires you to stare at it. Or an update that restarts your machine mid-session. Or an AI assistant that sends a message every time it takes a step, narrating its reasoning in real time when you would have been fine with a summary at the end.

The judgment call

The technical patterns are familiar enough: service workers, periodic background sync, optimistic updates. But the real challenge isn't implementation. It's the judgment about when to surface a result.

There is a temptation to over-notify. The tool did something. The tool found something. The tool is ready. Each of these is true, but announcing all of them costs the same 25 minutes. The respectful version asks: is this something the person needs to know right now, or something they can discover when they return to this context?

When an AI agent processes your notes and finds that a contact changed jobs, based on something you forwarded last week, it has two options. It can interrupt you now to ask if you'd like to update the person record. Or it can prepare a proposed change, queue it for the next time you open that person's page, and surface it then as a simple diff to review. The information is the same. The timing is different. And the timing is most of the respect.

This is the design question most AI tools haven't solved. They are built to be present, which feels like helpfulness but can easily become noise. The hardest part of building a calm tool isn't the background sync. It's knowing when silence is the right answer.

The test

Weiser died in 1999, before most of what we call AI tools existed. But the test he and Brown were pointing at is still the right one: does this technology move easily from the periphery of your attention to the center and back, or does it plant itself at the center and wait for you to deal with it?

Most tools fail the test by design. They are built to be noticed. They compete for your attention with everything else competing for your attention.

The rare tool that passes is the one you describe as "it just works." Which turns out to mean: it works without requiring you to watch it work.

---

Asgeir Albretsen is the founder of Harbor.