AI product
design.

AI features designed like any other feature. What the person asks, what they get back, and what happens when the answer is wrong.

A smartwatch on a wrist in daylight

What’s included

(05)

An AI feature that works in a demo can still confuse, mislead or frustrate the people who use it every day.

  1. The ask and the answer

    How people tell the feature what they want, and how the reply is shown so it is easy to check, edit or use.

  2. Failure states

    What the person sees when the answer is wrong, empty, unsafe or unsure, and how they recover without starting again.

  3. The wait

    Progress, partial results and a cancel button, so a slower AI feature still feels like it is working for you.

  4. Voice and tone

    Replies, refusals and languages that sound like your brand, not a generic assistant.

  5. Consent and a way to a person

    Plain words on what the feature does with people’s data, and a clear route to your team when the AI cannot help.

When you need it

  • You are adding a chat box or assistant and are not sure it should be a chat box.
  • Users do not trust the AI answers, or trust them too much.
  • Your AI feature works, but nobody uses it.
  • Support tickets are about the AI getting things wrong.

How it runs

(04)

You get a written scope, a fixed price and a start date before anyone begins.

  1. Frame

    We agree what job the feature does for the person and what a good answer looks like.

  2. Design

    Screens and flows for the ask, the answer, the wait and every way it can go wrong.

  3. Stress

    We run real and awkward inputs through it, including the ones meant to break it.

  4. Hand over

    Your engineers get the designs, the states and the rules, ready to build or ship.

What you leave with

  • Flows and screens for each AI feature
  • Designed failure, empty and unsure states
  • Loading and progress patterns for slow answers
  • Tone rules for replies and refusals
  • Consent copy and a human handover path

Start your project with us

FAQ

Answered questions.

The output changes every time and is sometimes wrong. So the design has to show how sure the answer is, let people correct it, and handle failure as a normal case rather than an edge case.

Often not. Many AI features work better as a button, a suggestion or a filled-in draft. We pick the form from what the person is trying to do, not from what is common.

Yes. We review what is there, find where people get stuck or misled, and redesign those parts. Testing on real devices is covered by our QA work.

Portrait in profile against a red ground