My AI-Powered Design Process

Before touching Figma, I use Claude to clarify intent, draft short PRDs, and surface risks, edge cases, and initial approaches. Faster alignment, fewer wasted cycles.

• Session recordings, NPS verbatims, and support tickets fed directly into context
• Patterns surfaced in hours, not weeks
• Every design decision traces back to real customer behavior

Talk directly with customers, run quick tests, and use what you learn to adjust direction. Research is a velocity tool, not a gate.

• AI synthesizes interview notes and identifies recurring themes
• Assumptions stress-tested before a single component is built
• Direction adjusts early, not after eng has already built it

No wireframes. Day one is a working prototype. Use Cursor and Claude Code to build and iterate on flows and simple interfaces. AI generates code while I guide structure, behavior, and UX quality.

• Clickable, interactive simulations from voiced prompts
• Stakeholders feel the product, not review a static mockup
• Partner with engineers to decide what moves into the product

Validated concepts move into Figma for full state coverage, system alignment, and production readiness. This is where craft happens.

• Accurate tokens, correct states, proper spacing, motion-ready
• Every component maps back to the design system
• Multiple form behaviors and flows tested simultaneously

No scheduled design crits. No screen shares. Prototypes go out async. Stakeholders review on their time, leave timestamped reactions.

• Syncs happen with real context, not cold first impressions
• Review cycles shrink, feedback gets broader
• Momentum stays

Enterprise products fail at the edges. I design for them on purpose.

• Empty states, permission boundaries, degraded network: all prototyped
• AI generates realistic data sets and simulates load behaviors
• Problems caught in prototype, not in production

Figma is the source of truth. Not “ready for handoff” in theory. Actually ready.

• Specs, tokens, states, and annotations structured for direct implementation
• In-browser reviews during build, flagging drift early
• Engineers implement intent, they don’t interpret it