How I work
Two diamonds. One AI-powered workflow.
The methodology stays constant — Double Diamond, always. What changes project-to-project is
the phase weighting. Domain expertise lets me compress discovery. MCP and Claude Code
accelerate build. Senior judgement is knowing where to speed up and where to slow down. I
design in Figma, Claude reads it via MCP and generates production code, the fidelity gap
collapses. AI as co-pilot, not autopilot — I choose when to use it, and where deterministic
logic beats LLM guesswork.
20% Discover
20% Define
35% Develop
25% Deliver
01 / Discover
Understand the problem space
Sit with the team and the users. Map constraints, workflows, the messy reality. This
phase doesn't accelerate — understanding the problem space takes the time it takes. AI can
summarize transcripts and pull patterns from research notes, but the conversations
themselves happen at human speed.
02 / Define
Frame the right problem
Synthesise what research actually says, not what everyone assumed it would say. Kill
scope that doesn't serve the core job. Sometimes this is a three-week reframe (Admiral
Money's pivot from form-conversion to quick-quote). Sometimes the brief is already sharp
and Define compresses to a week. Knowing which is senior judgement.
03 / Develop
Figma to production, via MCP
I design in Figma — components, tokens, layouts, flows. Claude reads the Figma file via
MCP (Model Context Protocol) and generates production-ready React code. No screenshots,
no manual handoff, no fidelity loss. The design file becomes the source of truth.
Engineers refine working components, not translating mockups. Develop is where AI
compounds hardest.
04 / Deliver
Stay close to launch
Pair with engineers through QA, polish the details that Claude missed, then watch it in
the wild. AI doesn't ship products. People do. The handoff still requires judgement calls
— what's MVP, what's polish, what breaks if we cut it. The gap between design file and
shipped product is smaller, but the last 10% still needs a human.
AI as co-pilot, not autopilot
Conversational AI, vision extraction, structured data generation — these are tools, not
replacements. I choose when to use them (Admiral Money's quote calculator: deterministic
math, not LLM guesswork). Senior judgement is knowing where AI accelerates and where it
introduces risk.