How I work with AI, one dated issue at a time. Older issues stay as written.
A design system that starts from the objects an operator works with, not from buttons and cards. Five axes, one grammar, checked on every build.
My designs leave as pull requests into develop, not Figma frames. The branches, the commands, the SIT loop with QA, and what happens when a developer and I touch the same logic.
What changed in agent work by Oct 2026: thinner scaffolding, ownership over job titles, and the check that says "not done yet."
Turning Figma files into production-ready frontend code through Figma MCP pipeline.
Multiple web properties, each with a dedicated UX sub-agent connected to Figma MCP write-to-canvas.
How a UX designer orchestrates AI agents through Claude Code. Not a framework. A real working setup.
How I build a design system in Figma with agents: brand guidelines and a component inventory become md specs, one agent draws from them, another checks against a rulebook.
Claude Opus 4.8 runs in both GitHub Copilot and Claude Code. Same model version, different harness — and when to reach for each.