The case for calm interfaces in AI products
By Allan Leone on
Shimmer borders, streamed text, sparkle icons. Most AI products are currently signalling their own cleverness on every interaction. That reads well in a demo and badly on the fortieth use of the day.
There is a house style in AI products right now that treats every response as an event. A shimmer on the border while it thinks. Text typing itself out at reading speed even though the whole response arrived at once. A sparkle icon, in case you had forgotten which part is the clever bit.
It works the first time. It signals capability to someone evaluating the product. The problem is that evaluation happens once and use happens daily.
The artificial streaming case
Streaming text made sense when it reflected reality. The tokens genuinely arrived over several seconds and showing them early let people start reading sooner.
A lot of implementations now hold a complete response and animate it out anyway, because it looks like thinking. That converts a fast interaction into a slow one on purpose, and users who read faster than the animation are left waiting for a machine to finish performing.
Novelty is a loan against attention. It comes due on the fortieth use.
What calm actually means here
Not undesigned, and not featureless. It means the AI is load-bearing without being loud, and every piece of motion is carrying information rather than signalling effort.
- Animate to show where something came from or went. Do not animate to fill time.
- Reserve the strongest visual treatment for the thing that needs the most scrutiny, which is usually the output the user is about to act on.
- Let fast things be fast. If the response is instant, show it instantly.
- Drop the badge. If the feature is good, the user does not need reminding which part was the model.
Why this is a commercial argument, not a taste one
Products in this category are moving from novelty adoption to daily use, and the qualities that win those two phases are different. Demo-friendly effects convert evaluators. Restraint retains users, because retention is decided by people who have already seen the trick and now just want the task finished.
Where to start
- Find every animation in your AI surface and ask what information it carries. Remove the ones where the answer is "it looks considered".
- If you fake streaming, try shipping the instant version to a portion of users and watch task completion time.
- Use your own product for a full working day. The things that irritate you by mid-afternoon are the ones costing retention.
- Move the visual emphasis to the review step rather than the generation step. That is where the user actually needs help.
Tags: ux, design, ai