martijnotter.com
↜ Back

How I Design with AI: Sketch, Validate, Iterate

This Isn’t “Vibe Coding”

Let me be clear upfront: this has nothing to do with vibe coding. I’m deliberate with my prompts and know exactly what I want to achieve. This actually fits perfectly into the design process. As designers, we spend our days figuring out how to solve problems and shape interactions. The deep thinking is already part of our job—and that’s precisely what makes designers well-suited to get great results from AI agents, rather than having them hallucinate during aimless coding sessions.

Before AI, my workflow looked like this:

  1. Spend hours brainstorming, hoping something would work
  2. Create static screens, hoping they’d actually function
  3. Spend days designing UX interactions and flows while building clickable prototypes (in recent years at least—before that it was pure guessing)
  4. Hand off to a developer
  5. Discover it doesn’t actually work technically
  6. Back to the drawing board, repeat everything

The New Way: Sketch, Validate, Iterate

Now, with AI, I use Figma or my Remarkable tablet as a sketchbook. Whenever I hear about a problem, I usually already have an idea in my head. My goal? Validate that idea as quickly as possible.

I don’t create complete designs in Figma anymore. Sometimes it’s just one rough sketch with a pencil. Then I jump straight into Claude.

My Process

1. Capture my thinking
I dump all my thoughts into Claude chat (just Claude desktop, not Claude Code). It’s hard for me to extract all the chaos in my head, so I’m training myself to just tell Claude directly. He can handle the noise and structure it, something I’m not good at.

2. Combine sketch + validation
By combining my sketches, the ideas in my head, and Claude’s validation, he helps me arrive at the right prompts. I add my knowledge of our codebase and specific components, and suddenly my design ideas are running in my browser.

3. Work in small steps
I never throw huge prompts at Claude. I know exactly what I want and tackle very small scenarios piece by piece.

I should mention: I know where to find things in our codebase and understand the files, so I can point Claude directly to the right places. During this process, I talk with Claude constantly, especially for technical problems where I need to understand exactly what the issue is and verify that I understand what a feature is supposed to do.

The Results

When I work this way, I end up with functional prototypes where everything actually works. The best part? While using them, I’m already thinking of the next feature, something you could never achieve by designing static screens.

That’s how I’m working with AI these days. Fast validation, real functionality, immediate iteration. The design thinking is still mine, AI just helps me validate and build faster than ever before.