A site and its design system, built from my specs
Gemini for desk research, Figma AI for wireframe variants, Claude for the build. Input: my IA map, content model, mood boards and token spec.
Read the case study →✦ How I use AI
AI does more of the making. I stay responsible for the judgment.
I use AI in research, exploration and building. It's only as good as the brief, so most of my effort goes into the research, structure and specs I give it, and into checking what comes back.
01
Where it fits
Each stage of my process, and who does it. The parts that need judgment about people stay with me.
| Stage | Me | ✦ AI |
|---|---|---|
| Research | Interviews, observation, synthesis | Desk research, summaries (Gemini) |
| Structure | Information architecture, content model | — |
| Direction | Mood boards, design system rules | — |
| Explore | Pick and refine | Wireframe and UI variants (Figma AI) |
| Build | Specs for every page and view | Front-end and prototypes (Claude) |
| Review | Check against research, IA and system | — |
02
Three examples
Planning, building and checking a relaunch with AI
Claude kept the task list and the deadlines, drafted pages from my notes, built the dashboard prototypes and flagged dates that didn't match between my CV and LinkedIn.
I set the scope, decided what to cut when time ran short, and checked every fact and every chart before it went live.
See the dashboard prototype →Designing for AI before designing with it
A model ranked Chile's mines by inspection risk. I designed the ranking to explain itself, which is why inspectors trusted it enough to plan their visits by it.
Read the case study →03
What I check
- 01 / Against researchDoes it answer the questions users actually asked?
- 02 / Against structureDoes every page sit where the IA puts it?
- 03 / Against the systemTokens, components and states, not look-alikes.
- 04 / Against realityReal data shapes, accessibility, and copy that says what's true.
✓ Where it helps mostTurning a precise spec into a working prototype in hours, so decisions get tested with realistic data instead of argued over mockups.
↺ Where I don't use itTalking to users, deciding what their answers mean, and deciding what matters. Those stay with me.