AI UI design is past the mockup stage: these interfaces work. We gave four frontier models three interface briefs, an analytics dashboard, a phone finance app and a kanban board, with fixed data and a fixed feature list, one attempt each, no edits. Every build below can be clicked, filtered, dragged and sent money through.
How did AI design a phone app?
Like a native app, with real feedback. All four finance apps put a bottom tab bar where the brief asked for it, respect the phone's top and bottom safe areas, and run the four-step send flow end to end: after sending $42.50, every app updates the balance from $4,826.19 to $4,783.69 and adds the transfer to the activity list.
- Claude Opus 5.5 drew a sparkline of the month inside the balance card, ended the flow on a full-screen green confirmation and is the only one that triggers the phone's vibration as haptic feedback.
- Claude Fable 5.1 added money-in and money-out chips and animated the new balance counting down under a "Sent!" overlay.
- GPT-6 Astra labelled each step 01 of 04, 02 of 04 in a quiet, airy layout.
- GPT-6.1 Sol drew an iPhone status bar reading 9:41 and confirmed with a slide-up review sheet.
The two GPT apps share a layout and the same typefaces, DM Sans with Manrope.
How did AI design a dashboard?
Correctly first, stylishly second. All four dashboards computed the same figures from the fixed dataset, $155,149.80 in revenue from 2,109 orders at $73.57 average, and all four support sorting, filters, search, pagination and a dark mode. The Claude dashboards add reading aids, a labelled peak and average line from Opus 5.5 and a best-day callout with a chart-to-table switch from Fable 5.1. The GPT dashboards lead with a friendly heading, "Good things are growing." and "Room to grow."
How did AI design an interactive tool?
With more care than you might expect. The kanban boards implement drag and drop with pointer events rather than the browser's built-in drag API, which is the choice a careful developer makes because it also works on touch screens. They show a placeholder where the card will land, confirm each move with an undo toast, keep your changes in local storage, and open card details in a side panel. Their dark modes are designed rather than inverted.
Which AI is best for UI design?
Pick by what the interface needs. For details that make an app feel finished, such as haptics, annotations and confident transitions, the Claude models led on every interface brief. For a clean, consistent interface in about ten minutes, GPT-6.1 Sol did the job at a fraction of the cost, with the caveat that it and GPT-6 Astra tend to design the same screen twice. For landing pages rather than product interfaces, see AI web design.
Four finance apps, names hidden. Which one would you trust with your money?
Take the blind test
Frequently asked questions
Can AI design a user interface?
Yes, and a working one. Four frontier models each designed and built a dashboard, a phone app and a kanban board from written briefs, with working charts, filters, a send-money flow and drag and drop, in one attempt and with no human edits.
What is the best AI for UI design?
For the most considered details, the Claude models: annotated charts, haptic feedback, richer transitions. For clean, consistent interfaces fast and cheaply, GPT-6.1 Sol. The two GPT models produced very similar interfaces, down to the same typefaces.
Is AI-generated UI accessible?
Largely, on our briefs. Every build carries ARIA labels and keyboard handling, and the finance apps all respect the phone's safe areas. Check colour contrast and focus states yourself before shipping, as you would with any design.
Does AI UI design replace tools like Figma?
It replaces the first working prototype, not the design process. The builds here are real code you can click, which is more than a mockup, but deciding what the product should feel like is still a human call.