Skip to content

Taste Is Trained

Taste gets talked about like a gift. Some people have the eye, and the rest of us install a component library. It is a comfortable story, because it excuses not practicing.

Nobody's sense of whether an easing is right came pre-installed. The people we call tasteful have noticed more things, more carefully, for longer. Taste is a skill, and it takes reps.

One of these cards has three small mistakes. Click anything that looks off, then reveal.

Found 0 of 3. Click anything that looks off.

If you found all three, good. If you found none, also good. You can only fix what you can see, and seeing gets better with practice.

The training loop

Emil Kowalski calls taste "a trained instinct", and his recipe is to surround yourself with great work, work out why you like it, and practice. In concrete terms:

Use good software slowly. Open the same menu three times. Ask what the shadow is doing, why the exit is faster than the entrance, what the hover state is not doing.

Name what you notice. "This feels nice" trains nothing. "The popover scales from its trigger, not its center" is a rep.

Rebuild it, then compare. Put your copy next to the original. The gap between them is the curriculum.

Change one thing at a time

Judging two options side by side is far easier than judging one on its own. Keep the component fixed and change a single variable. Press play a few times for each pair.

400ms
150ms

Once you can tell 150ms from 400ms without thinking, you have a reference point you did not have before. Do the same for easing, origin, shadow and spacing. Emil's Train your judgement is a whole page of pairs like these.

Why it matters more now

When a tool can generate a plausible interface in seconds, making one stops being the hard part. Picking the right one out of five is the job. As Emil puts it, AI "can write animation code. What it can't do is know what feels right."

Resources

Search concepts

Search for a command to run...