Concept.
Before.
Form.

I try to understand the problem before drawing anything. That usually means talking to users, reading whatever analytics and support tickets already exist, and looking at how people work around the current product.

AI helps with the synthesis: transcribing sessions, grouping notes, and surfacing patterns I can then check myself. It takes less time than it used to, and that time mostly goes back into questioning whether we are solving the right problem.

I prototype with Figma and Claude Code so people can click through something real instead of reviewing static screens. With AI, user testing became faster and more real.

A design system holds it together: tokens for colour, type and spacing, and components each screen inherits. That keeps the experience consistent, and it matters more when an agent is generating screens, because the system is what stops them drifting apart.

I build the real thing, put it in front of people, and change it based on what they actually do with it.

When AI writes part of the code, most of the work is in the setup. Each repo has a documented set of conventions the agent reads first, written test cases for the things that would matter if they broke, and types that reject anything the system never defined.