How I work
Context-intensive upfront. Execution-efficient downstream.
Research, strategy, user flows, and information architecture first, built as documentation in the repo with AI. Then design in code, with Cursor and Claude Code in the loop. Prototype, ship, validate, iterate on data. Figma stays in the toolkit when collaboration or design-system work needs it. Real user research stays in the toolkit when the problem needs direct insight. The point is the full loop: discovery to production to validation, all in one place.
Cursor · Claude Code · Antigravity · Figma · Next.js · Tailwind
Phase 1
Context and foundation
Before any design or code work begins, I build a comprehensive repository of context. Product requirements. User personas and journeys. Interaction models. Stack decisions. Component architecture. Content strategy. Success criteria.
AI tools produce better output when they have deep context. The prep phase is where the leverage compounds. Every downstream decision gets faster and more consistent because the constraints are already known.
Tools: Claude Code for repo building and documentation. Cursor for context-aware coding. Claude, ChatGPT, and Gemini in parallel for thought partnership. Different models surface different blind spots.
Phase 2
Design and exploration, tool-agnostic
Right tool for the job, not the other way around.
Figma when
- Team collaboration and design-system documentation
- Stakeholder presentations and alignment
- Complex visual exploration that benefits from rapid iteration in a canvas
- High-fidelity mockups for user testing
- Engineering handoff in a traditional team structure
Code when
- AI-native products that need real interactivity to be evaluated
- Rapid prototyping with production-ready output
- Complex responsive layouts across desktop, mobile, tablet
- Validating technical feasibility during the design phase
- Features that ship directly from the design work
Figma stays the collaboration language across teams. Code lets you validate behavior, performance, and accessibility immediately. Together they cover any team structure.
Phase 3
AI-assisted development
Production-ready front-end across desktop, mobile, and tablet. Design-system components built to industry standards. Accessible, responsive, performant. Across devices, not just at design time.
Full-stack capability when the project requires it (startups, solo work, side things). Deep collaboration with engineering when working in team structures.
Tools: Claude Code as primary environment, Cursor for fast generation, Antigravity as an alternate AI-native IDE.
Phase 4
Research and validation
Most upstream research and synthesis happens in the repo with AI. Competitive analysis, pattern identification, best-practice audits, technical feasibility checks. AI accelerates the work that used to take days into hours.
Real user research stays in the toolkit for when the problem needs direct insight. Interviews, usability testing, behavioral data. Some questions only users can answer. The method matches the question.
Working prototypes for testing, not static mockups. Real device validation. Metrics-driven iteration. A/B testing where it makes sense. Speed without validation is just fast failure.
The shape of the work has changed. The bar hasn't.
I'm still doing the same job. Understanding users, framing problems, shaping product behavior. AI doesn't remove the design work. It changes where it shows up. The thinking happens earlier and the artifact happens later, and most of the artifact is the running thing.