I'm Not a Designer. So I Let AI Design My App.
Kirill Ushakov - Web & Mobile Development
0:00 / 0:00
I'm Not a Designer. So I Let AI Design My App.
119 просмотров · 12 дн. назад
Kirill Ushakov - Web & Mobile Development
536 подписчиков
119 просмотров · 12 дн. назад
GitHub: https://github.com/kir-ushakov/public...
What happens when a solo developer needs a designer?
In the past, my answer would probably have been simple: hire one.
I'm not a UI designer, and the first version of Brain Assistant made that pretty obvious.
But AI changes this part of software development too.
In this video, I show how I'm using ChatGPT and Cursor not just to generate individual mockups, but to build a design workflow that gives AI enough context to keep the UI consistent as the application evolves.
Inside the codebase, I keep the design context under docs/design.
A UI_STYLE_GUIDE.md defines the visual rules and direction of the application. A structured design-system.json turns those rules into exact values and design tokens. Reference images show what existing screens actually look like, while JSON specs describe their structure, components, states, and behavior.
Together, these files give AI both visual and structural context when it works on the UI.
Then I put the workflow to use.
I show how I create a completely new UI state for adding a tag with a very short prompt in ChatGPT, using the existing screen, its spec, the style guide, and the design system as context.
I also take an old login screen that I designed myself and ask ChatGPT to redesign it to match the rest of the application — with almost no visual direction beyond the existing design context.
Once I'm happy with a mockup, I generate its spec and pass both to Cursor for implementation.
We cover:
• Keeping UI design context directly inside the codebase
• Using a style guide to define visual rules for AI agents
• Turning those rules into exact values with a structured design system
• Using reference images as visual context
• Describing screens, components, states, and behavior with JSON specs
• Connecting screen specs to their relevant visual references
• Generating new UI states with very little prompting
• Redesigning existing screens with ChatGPT
• Passing mockups and specs from ChatGPT to Cursor for implementation
• Delegating most of the actual UI design work to AI while keeping the overall direction consistent
The interesting part for me isn't that AI can generate a nice-looking screen.
It's that once the design context is structured properly, I can reuse it across ChatGPT and Cursor instead of explaining the application from scratch every time.
I'm definitely still not a designer. But for this kind of work, I'm starting to need one a lot less.
In the next video, I'm moving back to engineering — and to one of the fundamental problems in any offline-capable PWA: keeping local data synchronized with the server.
This series follows Brain Assistant as a real-world experiment in AI-assisted solo software development.
Can AI actually multiply what one experienced developer is capable of — enough to build and maintain a serious production product alone?
That's what I'm trying to find out.
#AI #AICoding #UIDesign #AIDesign #SoftwareEngineering #SoloDeveloper #SoftwareDevelopment #ChatGPT #Cursor #AgenticCoding