AI-generated interfaces vs. a designed product
This is not a page about whether AI can design. It is a page about the distance between a generated interface and a product people trust, and who closes it.
The short answer
A generated interface is the fastest way to find out whether anyone wants the thing. It gets you to real users in days, and for a prototype or an internal tool it is often where the work should stop.
It stops being enough at the point where someone outside your company has to rely on it. Not because it looks bad — it usually looks fine in a screenshot — but because the parts a generator does not draw are the parts that decide whether the product feels trustworthy.
We are not arguing against the tool. We use the same tools daily, and we say exactly where in a separate piece. We are arguing that generating and finishing are two different jobs, and only one of them is done.
Side by side
| A generated interface | A designed product | |
|---|---|---|
| Hierarchy | Everything on the screen looks equally important | One thing earns the eye first, and it is the right one |
| Empty and error states | The happy path, usually | Drawn for the cases that actually happen |
| Edge cases | What you thought to ask for | Four hundred rows, a very long name, a failed payment |
| Consistency | Each new screen is its own decision | A system that settles the decision once |
| Accessibility | Whatever the component library defaults to | Contrast, focus order and keyboard paths checked |
| Handoff | Screens, and a prompt history | Tokens, components and rules an engineer can build against |
| Your brand | Looks like the tool that made it | Looks like you |
| Where it is enough | Prototypes, internal tools, finding out if anyone wants it | Anything a customer pays for or a procurement team reviews |
When each one is the right call
Keep generating
When generated is genuinely enough
You are still finding out whether the idea works. The users are your own team, or a handful of friendly design partners who forgive rough edges. Every week of design is a week not spent learning something. Ship it and keep going.
Do not hire anyone yet. Including us.
Bring in a studio
When it needs designing
Real customers are paying, or about to. Support keeps hearing the same confusion. Sales cannot demo one particular screen. Every new feature restarts an argument about how it should look, because nothing settles it.
This is the work we do.
Questions we get
Are you going to make us start over?
Almost never. What you shipped is the most valuable research you own, because real people have used it and told you where it hurts. We keep the product and fix what is actually wrong, which is usually hierarchy, states, edge cases and the absence of a system rather than the feature set.
You use AI yourselves. Is this not a contradiction?
No, and we would rather say so plainly. AI is our daily driver: it explores twenty directions in the time three used to take and it scaffolds the code underneath a build. What it does not do is decide. It has no taste of its own and no stake in whether the work lands, so a person makes every call that matters and the work is finished by hand. If the output reads as generated, it is not done.
Can our engineers keep building after you leave?
That is the test we design for. A system that only works while we are in the room fails the first sprint after we leave, so the deliverable is tokens, components and the rules that hold them together, documented well enough that a screen we never drew still looks like ours.
What does it cost?
Engagements start at $6,000. That is our floor for any project rather than the price of a product engagement — a first one is usually a multiple of it. You get a fixed number before you commit.
How long does it take?
Weeks rather than months for a first engagement. Most product work then continues as an ongoing relationship, because a design system needs maintaining as the product grows.
Do you only work on AI-built products?
No. Most of the work is ordinary product design for teams whose software has outgrown the version that got them here. An AI-built first version is simply the most common way a product arrives at that point now.
Send us the thing you shipped.
However rough. We will tell you what is actually in the way of it feeling finished, and whether it is worth paying anyone to fix yet.