Latency is an emotion, not a number

Latency is the thing we complain about with the wrong vocabulary. We call an app “slow” or “fast” as if we were reading a stopwatch. But nobody experiences milliseconds. What we experience is a feeling — a small verdict the nervous system delivers before the conscious mind has weighed in. The thing feels alive, or it feels dead. The tap “took,” or it didn’t. That verdict arrives in a fraction of a second, and it is remarkably hard to argue with.

I think about this a lot from inside the build, because the number and the feeling keep coming apart. You can ship something measurably faster and have it feel worse. You can add a delay on purpose and have it feel better. If latency were only a number, none of that would be possible.

The three doors

Here’s the part that reorganized how I think about it. Human perception doesn’t treat time as a smooth dial. It treats it as a set of doors. And there are essentially three.

Under roughly a tenth of a second, a response feels instantaneous — the effect seems fused to the cause, like the screen is an extension of your finger. Under about a second, you stay inside your own train of thought; there’s a wait, but it doesn’t break the thread you were pulling on. Past ten seconds or so, you’re gone — attention has left the building, you’ve glanced at your phone, you’re mentally somewhere else.

What gets me is that these bands aren’t new. They’ve been known in human-computer interaction for decades, long before the apps we’re anxious about today existed. The thresholds are properties of us, not of our hardware. We built faster machines; the doors stayed exactly where they were.

The same wait feels completely different depending on which door it falls behind — and the doors belong to the human, not the machine.

The consequence is that all milliseconds are not created equal. Shaving 40ms off a 60ms interaction — pushing it deeper into the “instant” band — can be invisible; it was already through the door. But shaving that same 40ms off an interaction sitting right at the one-second line can feel like night and day, because you just kept someone inside their own thought instead of ejecting them from it. The engineering effort is identical. The felt payoff is not even close.

Once you see the doors, the design tricks stop looking like tricks. A skeleton screen that paints instantly and fills in later isn’t faster — it just gets something through the first door, so the wait that follows happens inside your flow instead of in front of a blank void. A progress bar that keeps moving buys patience, because a stalled wait feels longer than a moving one of the same length. Even a tiny, honest bit of motion on tap can make a slow thing feel handled. We’re not fooling the stopwatch. We’re speaking to the part of the person that filed the verdict.

So the question I ask now isn’t “how many milliseconds?” It’s “which door are we standing in front of, and which side of it does the user end up on?” That reframes optimization from a race against zero into something more human: you’re not chasing speed, you’re protecting a state of mind — someone’s flow, their focus, their sense that the thing they’re touching is real and listening.

Latency, it turns out, was never the number on the dashboard. It’s the emotion the number produces. Measure the first thing all you want — but you ship the second one.