My AI-powered design-to-code workflow

People often ask whether AI is replacing designers and developers. In my day-to-day work it’s doing something more useful: it’s removing the grind between an idea and a tested interface. Here’s the workflow I actually use, step by step, and the parts I never hand over.

1. Research: AI as my note-taker

After user interviews I’ll have pages of messy notes. I give them to an AI assistant and ask it to cluster the feedback into themes, pull out direct quotes and list open questions. What used to take an afternoon now takes twenty minutes.

What I don’t do is let it tell me which problem matters most. That comes from being in the room, hearing the pause before someone says “it’s fine, I guess”.

2. Ideation: generate ten, keep two

AI is a brilliant sparring partner for flows. I describe the user’s goal and ask for alternative journeys, edge cases and failure states. Most suggestions are average, but one or two are angles I’d have missed. Quantity is cheap now, so I use it and then edit hard.

3. Design: AI for the boring parts

In Figma I use AI for realistic placeholder content, quick layout variations and first-pass image concepts. It keeps me out of “lorem ipsum land”, where designs look great until real content breaks them.

AI can fill a screen. It can’t tell you what the screen should make someone feel.

Visual hierarchy, spacing, the design system and the tiny details are still hand-crafted. That’s where the real craft lives.

4. Build: pair programming, not autopilot

This is where AI saves me the most time. I use AI coding assistants to scaffold components, write tests, refactor and explain unfamiliar code. I treat every suggestion like a pull request from a fast but junior teammate: I read it, question it and make it accessible before it ships.

Scaffolding: first drafts of components from a clear description.
Tests: generating cases I then check and extend.
Refactors: repetitive changes across many files.

5. Test: humans only (mostly)

AI can draft usability test scripts and summarise session recordings, but the sessions themselves are with real people. Nothing replaces watching someone hunt for a button you thought was obvious.

What I never hand over

Deciding which user problem is worth solving.
Final visual and interaction decisions.
Accessibility sign-off.
Any code I don’t understand.

AI makes me faster. Judgment is what makes the result good. I’ll keep sharing what works and what doesn’t as the tools evolve.