With a human in charge
AI for design teams
I use AI every day to build things for design teams: tools that keep AI-generated code on-brand, check builds against Figma, and take the admin out of design work. AI does a lot of the typing. I make the decisions.
I've been working with AI since well before the current wave of hype. As chief product officer at a behavioural biometrics start-up, I directed the teams that built our machine learning models. These days I build with Claude Code, Figma and the Model Context Protocol, and I train design and product teams to do the same.
Projects
- Making Willowmere: a portfolio game built with Claude Code (This site): A pixel-art portfolio game built part-time, with Figma Weave for the art and Claude Code for the build.
- UX Tickets: Azure DevOps tasks on the Figma canvas (Figma widget): Azure DevOps tasks on the Figma canvas: tick a task and the real ticket closes, without leaving Figma.
- UAT: checking live builds against Figma with AI (Claude skill): Checks a live page against its Figma design at every breakpoint, then a person reviews the report.
- Design Systems for Humans: a design system with its own MCP server (Web app · MCP): A design system with its own MCP server, so AI coding tools build with the real tokens and components.
- Icon and Type Foundry: icon fonts and typefaces from Figma (Figma plugins): Icon fonts and installable typefaces, built from your Figma drawings.
- Claude skills for design teams (Claude skills): Ways of working written down once, so Claude follows the same standards every time.
- A GPT tour finder for On the Go Tours (Client project · GPT): Tell it what you like and when you can travel, and it suggests a tour and shows you the itinerary. Built in early 2023.
- Free Claude skill: write better UX user stories (Free Claude skill): A free Claude skill that writes Jobs To Be Done user stories for Azure DevOps, Jira, Trello, Notion and more.
How I work with AI
- One task at a time, with a written brief. My Claude skills are those briefs, written once and reused.
- Real sources, not guesses. AI tools get my design tokens and components through an MCP server, so they can't invent a new shade of blue.
- A person signs it off. The UAT skill writes the report, but a reviewer accepts it before it goes anywhere.
Common questions
Can AI coding tools use our design system?
Yes, if you give them a way in. An MCP (Model Context Protocol) server lets tools like Cursor, Claude and VS Code ask for your real design tokens and components while they write code. Without one, they guess, and they guess differently every time. Design Systems for Humans includes one.
Can AI check a build against the Figma design?
It can do most of the measuring. My UAT skill compares a live page with its Figma frame at desktop, tablet and mobile sizes, checks spacing, copy and colour against your tokens, and grades what it finds. A person still reviews the report and decides which differences are worth fixing.
Will AI replace designers?
Not the ones who can design a path through it. Most AI tools are a handful of language models with a different interface on top, which makes AI a UX problem. I've written more about this in how to upskill your UX team on AI.
How do you train a design team on AI?
I start by having teams critique AI output against their own design system and research, then write their rules down as briefs the AI has to follow, then pair confident people with nervous ones on live work.
Working with me on AI
At Designed for Humans I run AI training for design and product teams, and I set up AI coding tools to follow your design system using Design Systems for Humans. See how we can work together or book a call.