Latency Showdown: The 10ms Coffee Challenge

Humans are strange creatures. We will wait patiently at a DMV for three hours,
but if a website takes more than 300ms to load an image, we start muttering proclamations
of doom like “this site is broken” or “my internet sucks.”

So today’s experiment asks a silly but revealing question:

How much faster does an interaction need to be
before a regular person feels the difference?

To test this, I built a tiny mock API called the Coffee Machine, with 3 latency modes:

Baseline: ~180ms
Slightly optimized: ~140ms
Over-optimized: ~10ms

The Setup

The fake API randomly returns “☕ Latte ready!” but delays depending on scenario.

export async function getCoffee(mode: "baseline" | "mid" | "fast") {
const delay = mode === "baseline" ? 180 :
              mode === "mid" ? 140 : 10;

await new Promise(r => setTimeout(r, delay));

return { message: "☕ Latte ready!", delay };
}

The API isn’t real, obviously.
The point is to simulate a real perception threshold experiment.

Results

Using the mock latency chart in the hero section of the site, I ran 100 simulated requests:

p95 Latency · Baseline (no caching)

p95 Latency · Cached

p95 Latency · Pooled connections

The funny part:

Almost nobody noticed the drop from 180ms → 140ms
Everyone noticed 140ms → 10ms
But many couldn’t articulate why they felt it was better — they just “liked it more”

This supports a classic UX principle:

Small latency wins don’t matter. Big wins do.
And humans notice speed emotionally, not logically.

Reflection

This experiment isn’t about coffee.
It’s about understanding where optimization time actually pays off.

Shaving off 10–20ms from a UI interaction is cool, but it won’t make your product magical.
Shaving off all but 10ms will.

Sometimes the most important optimization
is simply the one a human can feel.