How I Use AI in My Design Workflow

What's Actually Changed

A year ago I was skeptical. AI-generated UI looked like UI from a stock photo library — technically correct, spiritually empty. It couldn't understand context, constraints, or the weird edge cases that make real product design hard.

That's still largely true for visual generation. But AI has genuinely changed three parts of my workflow.

Writing First Drafts

I use LLMs to write first drafts of UX copy — error messages, empty states, tooltip text, confirmation dialogs. I don't use the output directly. I use it to see what a reasonable answer looks like, then rewrite it to match the product's voice and the user's mental model.

The time savings are real. More importantly, it forces me to have a reaction: "that's wrong because..." which is often faster than generating the text from scratch.

Exploring Problem Framing

When I'm stuck on a design problem, I'll describe it in plain language and ask for alternative framings. Not solutions — framings. "What are different ways to think about the relationship between a video and its watch options?"

It's like rubber duck debugging but the duck has read a lot of design books. Occasionally it surfaces a frame I hadn't considered. More often it confirms I was already thinking about it the right way.

AI-Assisted Review at Work

The most concrete impact I've seen is in operational workflows — specifically, using LLMs to pre-review structured content (banner ads, metadata, localization strings) before human review. This was a direct work project.

The insight: AI is excellent at rule-based compliance checking when the rules are codified. It's not good at judgment calls. Design the workflow so AI handles the former and humans own the latter.

What Hasn't Changed

Design requires understanding who you're designing for. No AI tool I've used can replace 30 minutes watching a real user try to accomplish a real task. That's still where the real information lives.