AI Builds Pages. Designers Build Products.

I've been building with AI coding tools for the last eight months.

Not demos and prototypes — real apps.
Revenue-generating systems.
Security-audited products.

The speed is absurd.

You can go from idea to working software in hours.
Architecture, components, dashboards, tables.

It feels like cheating.

AI has removed engineering bottlenecks.
It has collapsed iteration time.
It has reduced the cost of production to almost zero.

That includes design production.

Layouts are cheap.
Exploration is cheap.
Polish is cheap (well, cheap enough).

But cheap execution does not eliminate judgment.

Here's the distinction I keep coming back to:

AI is good at UI.
It is not very good at UX.
Yet.

UI shows users data and actions.

UX shows users the right data
and the right action
at the right time.

That guiding layer — prioritization, sequencing, hierarchy — is where senior product designers live.

AI doesn't naturally operate there.

It builds page by page.

Design builds systems.

That's why so many AI-built apps feel “off.”

They look modern.
They feel clean.

But something's wrong.

Buttons change color between screens.
Hierarchy shifts.
Badges are rounded here and square there.
Status colors are arbitrary.
You get multiple labels for the same data.
Similar data is treated visually differently.

There isn't a theme. Just the concept of a theme.

And then there's the real trap.

There's a principle in UX research called the Aesthetic–Usability Effect.

When something looks beautiful, users assume it's easier to use.

AI makes beautiful interfaces fast.

So the first impression is strong.

“This looks great.”

Expectation is set.

But if the structure underneath is incoherent — if the flow is out of order, if onboarding overloads you, if the primary action isn't clear — frustration is amplified.

Because now it feels broken.

Polish creates a promise.

UX has to fulfill it.

This is where AI struggles most: flow.

It doesn't instinctively know how much belongs in each step.
It doesn't model dependency trees well.
It will ask for Step 3 before Step 1.
It will drop a critical workflow into a tiny modal because that's the default.

It can build the pieces.

It doesn't always know which piece should come next.

That's not a knock on AI.

It's a reminder.

AI is a collaborator.

It is your super genius intern, overflowing with ideas and enthusiasm, with absolutely zero experience to guide decisions.

You wouldn't ship AI-generated backend code without reviewing it.

You shouldn't ship AI-generated UX without refining it.

AI can build almost anything now.

But it still can't decide what matters.

Products are defined by what matters.

That's the work.