Designing with AI, not around it
AI didn't replace my craft — it changed which parts of it matter most. A working designer's notes on weaving generative tools into a real product workflow.
When generative AI showed up in my workflow, the first thing it did was make me defensive. If a model can produce a passable login screen in three seconds, what's left for the designer? Two years in, my answer is: almost everything that matters. The work just shifted, and a few muscles I had been quietly practicing — research synthesis, problem framing, taste — became the entire job.
What AI is genuinely good at
Speed of variation, mostly. Asking for ten layouts of an empty state, fifteen tones for an error message, or three competing IA proposals used to be a full afternoon. Now it's a coffee. The work isn't more original than what I would have made — it's just faster to triangulate towards the right shape.
- Research synthesis: clustering interview transcripts in minutes, not days.
- Copy and tone exploration: testing voice across an entire flow before pixels exist.
- Boilerplate craft: forms, settings pages, table densities, dark-mode variants.
- Accessibility passes: catching contrast issues, missing labels, vague link text.
Where it falls flat
AI does not understand context that hasn't been written down. It cannot feel the seven unspoken constraints from your last steering committee. It cannot remember the engineer who asked, three sprints ago, that we never use a left-aligned modal again. The model is fluent — it is not informed.
"AI gives me more drafts. The job has always been deciding which draft to ship."
How my workflow actually changed
I split my day into two modes. In divergent mode, I pair with the model — generating, riffing, throwing things away. In convergent mode, I close the model entirely and look at the work the way a skeptical user would. Mixing the two is what kills good work. The model wants to keep producing; the designer's job is to stop.
Three principles I work by now
- Treat AI output as the rough draft, never the final argument. The model proposes; I dispose.
- Spend the saved time on research, not on more screens. The cheapest hour is generating; the most valuable is interviewing.
- Keep the brand voice in your head, not in the prompt. If you outsource taste, the work loses its accent.
What I'd tell a junior designer
Learn the model — properly. Learn its blind spots faster. The designers who will struggle in the next five years are the ones who treated craft as the whole job. Craft is the floor now, not the ceiling. The ceiling is judgment: knowing what problem to solve, what to leave on the cutting room floor, and when the elegant solution is also the wrong one.
AI didn't make me a worse designer. It made the parts of me that were already a designer matter more.