Notes on designing with AI

A living list of my observations/discoveries from using AI in my design workflow.

• AI-assisted prototyping as a starting point. I start most new projects by getting AI to prototype my first idea. I get to the edges of a problem faster.

• Code-only AI workflows still trap you. You end up in the code/screenshot/iterate loop (just now with an AI to wrangle) and risk locking in and optimising the current version too early. Comparing different working options is also awkward ("Undo that last change ... no not that bit, the other bit from earlier").

• Figma still has a role to play. After the AI-coded prototype, I go to Figma. Exploring quickly and seeing lots of different options together sparks new ideas in a way that code doesn't.

• File-based context is fragile. Using skills.md, reference.md, etc. for offloading research analysis or writing specs makes a patchwork of easily-lost folders and files, susceptible to the usual sync/“Where did I put that?” dance.

(Updated: April 2026)