Prototypes close enough to the real thing that the team builds from them. AI is how I get there quickly.
The workflow
User journey design
SME interviews, screenshots of the legacy system and policy documents pulled together into a journey that makes sense, with AI doing the first pass at the synthesis.
Feedback analysis
Research sessions recorded and transcribed, personal data taken out, then the transcripts questioned with AI until the useful findings come out.
Rapid prototyping
Complex, interactive prototypes built much faster with Claude Code and Codex: in the GOV.UK Prototype Kit, and from scratch in Dynamics 365 when the service runs there.
Prototype as source of truth
The prototype replaces most of the static documentation. Business analysts write acceptance criteria from it in refinement, and developers read the logic straight from it when they build.
Building my own tools
When a new model came out I used it to write a Figma plugin that builds a Figma page from a prototype screen. A side project rather than team process, but it worked, and it is the kind of thing I will try when the tool I want does not exist yet.
Automated documentation
Commit messages, pull request descriptions and change notes written as I go, so whoever picks up the codebase next can see what changed and why.
Picking up a new stack
AI to get productive quickly in something unfamiliar, React or Dynamics 365, so I can prototype in whatever the service is built on.
The tools
- Discovery and synthesis
- The latest AI models
- Summarising research, analysing feedback, drafting journeys and mapping conditional paths
- Prototyping
- Claude Code and Codex
- GOV.UK Prototype Kit, accessible HTML and Nunjucks, and Dynamics 365 built from scratch
- Design tooling
- Figma
- Screen flows and designs for the team to review
- Documentation
- Claude Code and GitHub Copilot
- Commits, pull request descriptions and change notes
- Image generation
- Gemini and ChatGPT
- Imagery for prototypes, decks and this site