AI does the making. I do the deciding.
I use AI every day, from the first messy brief to working code. It turns days of production into hours, which leaves more time for the part that matters: deciding what to build and making sure it's right.
How I think about it
Speed is for learning
Fast prototypes exist to get real feedback sooner, not to skip thinking. A testable concept on day two beats a polished deck on day ten.
Judgment stays human
AI can generate options. Choosing between them, and knowing which ones to throw away, is still design work, and it's mine.
Everything gets checked
Every output is reviewed like a junior designer's work: for accuracy, accessibility, consistency with the system, and whether it actually solves the problem.
The tools I reach for
Each tool has a job. None of them has the final say.
Thinking partner
Claude
Synthesizing research, pressure-testing problem statements, drafting flows and microcopy, and critiquing work before it reaches a review.
Design system work
Claude with Figma MCP
Reading and writing directly in Figma: auditing components, generating variants and states, and keeping tokens and specs in sync.
From design to code
Claude Code
Turning Figma designs into working front-end, so prototypes run in a real browser. This site was built this way.
Interactive concepts
Figma Make
Quick, clickable concepts inside Figma when an idea needs to be felt rather than described.
Testable prototypes
Lovable
Full, data-backed prototypes that users can actually try, for research sessions and stakeholder demos.
Working with engineers
Cursor
Reading and adjusting code alongside the team, so design changes land in the product and not just in the file.
Where AI helps, and what stays human
Frame the problem
Where AI helps
Pressure-tests the brief, lists assumptions and open questions, drafts success metrics to argue with.
What stays human
Agreeing with the team on what we're solving and what success means for the business.
Discover
Where AI helps
Clusters interview notes and survey answers, maps competitors, surfaces patterns across sources.
What stays human
Running the conversations, and spotting the one insight that reframes everything.
Synthesize and decide
Where AI helps
Turns findings into options, drafts the brief, writes up the directions we're not taking.
What stays human
Making the call, and owning it.
Prototype at speed
Where AI helps
Builds clickable and coded prototypes in hours with Figma Make, Lovable and Claude Code.
What stays human
Deciding what's worth prototyping, and what the test needs to prove.
Design and build the system
Where AI helps
Generates component variants and states through Figma MCP, checks tokens, writes specs.
What stays human
Craft, motion and edge cases, and keeping the system coherent as it grows.
Ship and measure
Where AI helps
Helps pull together analytics and feedback into a readable retrospective.
What stays human
Reading what the numbers mean, and sharpening the next brief.
This website
I designed this portfolio in Figma, then built it with Claude: reading the file through Figma MCP, writing the code with Claude Code, and refining it through conversation, one round of feedback at a time.
Claude handled the production: layout, responsive behaviour, animation, accessibility checks and the assistant you can chat with in the corner. I handled everything that needed taste or a decision, from the art direction and copy to which ideas to cut.
Claude did
- Turning the Figma file into code
- Responsive layouts and motion
- Accessibility and performance checks
- The chat assistant
I did
- Art direction and visual design
- Content, positioning and copy
- Every review and every call to change course
- Knowing when it was done
The rules I keep
No confidential client data goes into an AI tool without the client's approval.
Facts, numbers and research findings are checked against the source, every time.
Generated UI is reviewed against the design system and accessibility standards before anyone sees it.
If I can't explain why a design decision is right, it isn't finished, whoever drafted it.
Want to see it on a real problem?
I'm happy to walk you through a project and show how AI fit into each step, including where it didn't.