BUILDING
design
systems
WITH AI

My approach depends on what already exists. If a system is in place, I use AI (Claude) to extend it while staying consistent. If nothing exists yet, I use AI to quickly establish the foundation and build documentation, components, and code together. Both paths are shown below.

A 3D animated depiction of a two-layer LEGO brick, with a red bottom layer and a purple top layer with four studs, floating in mid-air against a gradient blue background.
Path A

A design system already exists

Goal: add a new component without drifting from what's already there.

  • Ideate the pattern
    A designer sketches the new component in Figma — by hand, or by prompting Figma Make for first-pass variants.
    Figma
  • Pull it into Claude
    Claude reads the new frame through the Figma MCP server — layers, variants, auto-layout, and which tokens it uses.
    Figma MCP
  • Check it against the system
    Claude reads the design-system docs for naming conventions, token names, and accessibility rules that already apply.
    Docs MCP
  • Match existing code patterns
    Claude reads the component library in the repo — how similar components are structured, prop naming, test conventions.
    Repo MCP
  • Generate the component
    Claude writes code that reuses existing tokens and primitives, then opens a branch and a pull request.
    ClaudeRepo MCP
  • Update the docs
    Claude drafts the usage guidelines, props table, and links back to the source Figma frame.
    Docs MCP
  • Review and merge
    Design and engineering review the PR together. Once merged, the system has grown by one component — fully wired end to end.
    Repo
Path B

No design system exists

Goal: stand one up quickly, from whatever fragments already exist in the product.

  • Audit the code
    Claude scans the product repo to reverse-engineer the colors, spacing, type, and recurring UI patterns already shipped.
    Repo MCP
  • Audit any loose Figma files
    If design files exist anywhere, Claude reads those too. If not, this step is simply skipped — code is the source of truth.
    Figma MCP
  • Ideate the foundations
    Use Figma Make to rapidly prototype a small core set — button, input, card — from the audited patterns, or design them by hand.
    Figma
  • Define the tokens
    Claude proposes a token set — color, type, spacing, radius — from the audit and new ideation, and writes it up as the first docs.
    ClaudeDocs MCP
  • Scaffold the codebase
    Claude sets up the component library structure — folder conventions, build config, first components wired to the new tokens.
    Repo MCP
  • Build the loop
    For each next component: ideate → Claude reads Figma → checks docs → writes code → docs updated. All three grow in lockstep.
    Figma MCPDocs MCPRepo MCP
  • Add governance
    Claude drafts contribution guidelines and a sync check so Figma, docs, and code don't drift apart as the system scales.
    Docs

UX
Process

I use the Double Diamond to cut through ambiguity and make high-impact decisions quickly, aligning user needs, business goals, and system constraints to deliver scalable, production-ready solutions. This approach allows me to identify the right problems early, reduce unnecessary iteration, and ensure what we build is both effective and sustainable at scale.

What
Guides
My
Design