AI agents
Design systems
Research
Prototyping
Tools
Claude, Claude Code, Figma MCP, Framer
Used for
Design system, research, prototyping, this site
Built
A design-system agent and Federico
Rule
AI does the repeatable work; design decisions stay with designers
6 → 2
UX designers in under a year, with no design system
144
AI-moderated interviews after the Webjet launch
Daily
How often both designers use the agent
A design system for a team of two
Why I built it: In under a year, Webjet’s UX team went from six designers, including a dedicated researcher, to two. It happened through two leadership changes, right after a rebrand that launched without a design system. I now lead the work of building one, but there isn’t a team to build and maintain it the traditional way, so I built an agent to carry the design rules while we do.
What it does: It works from the brand guidelines our agency, The Hallway, gave us and the components we built for the rebrand pages. I built it with Claude Code, a Claude Project and the Figma MCP. It generates on-brand cards directly in Figma: the kind of repeatable components that used to need a designer for every variation.
Checking against the rules: It also reviews designs against our existing mocks and rules. If a card uses 16px padding in one flow and 14px in another, it flags the mismatch and turns it into a list of questions for me to resolve. Nothing reaches developers without a designer’s decision.
Annotation, designed with developers: Annotating designs for handover used to take up a lot of our time. I asked our developers how they wanted designs annotated for build and set the agent up around their answers. It now annotates its own output in the format they asked for.
What changed: I use it every day, and the other designer now does too. Developers get build-ready designs sooner, with fewer review rounds and less waiting on us. That frees me for quarterly planning, strategy and the design work that needs judgement. New patterns and trade-offs still go through us.
001
The agent at work. A demo recreation with a sample brand, sped up. It builds a card from a one-line brief, checks it against the brand rules and raises issues as questions, then annotates it for handover.
Research, with AI in the loop
What we did: After the Webjet rebrand, our researcher and I ran AI-moderated interviews with 144 travellers in Australia and New Zealand, all within the first month after launch.
How I used AI: I used Claude alongside Dovetail and Hotjar to work through the results, pulling out themes and checking what people said against what they actually did on the site.
What stayed human: Deciding what mattered. The AI surfaced patterns; the conclusion that price and fees still decided bookings was ours, and the personalised add-ons pitch that came from it was mine.
Why it matters: AI moderation let a small team hear from far more travellers, far sooner, than we could have interviewed ourselves.
002
Building with AI
Prototyping: I use AI to prototype, debug code and generate realistic content and edge cases, so I can test ideas against something close to real much earlier.
Federico: The cat on this site is a chatbot I built. He knows my work and answers questions about it, in character. Ask him anything.
Side project: I’m building an AI orchestration project to learn how far these tools can go.
003
Federico. The chatbot I built for this site, answering a question about my design-system agent.
Bringing the team along
What I run: Workshops to bring my team up to date with AI tools.
What they cover: Where the tools genuinely save time, and how to fit them into how we already work.
Sharing what I build: I handed the design-system agent to the other designer, and she now uses it every day too.
Why it matters: With a smaller team, everyone needs leverage. Less time on repeatable work means more time for the decisions that need people.
004
Where I draw the line
Decisions stay with designers: AI speeds up how I get there, but it doesn’t decide what’s right for the people using the product.
A designer signs off: Nothing the agent produces reaches developers without a designer’s decision.
Built around the people who use it: The agent’s annotations follow what developers told us they needed, not what I assumed.
Leverage, not a replacement: With two designers, AI is how we keep quality up. It doesn’t replace the research, conversations and judgement the work depends on.
005