Minimum lovable beats complete. With my PM I cut every feature down to the smallest version people will actually love, and ship that first. I once spent a couple of sprints on per-location controls almost nobody used. The hard part is knowing what to cut.

The design isn't finished at handoff. Before we build, I mark which events we need to track in BigQuery. After release I watch session replays, read the feedback, and write to users directly. One small modal I added when people turn a feature off has brought in 4,850 feedback messages in a year, and a lot of what we shipped next came straight from them. For me a design is done when the users, my gut feelings and the analytics tell it works.

I work with engineers, not above them. At grooming I explain what the feature is and why we build it, and I want engineers to question it. During the build I'll change things to make their work easier, as long as it doesn't hurt the UX. I help the team through development and QA. I don't just hand over the mockups and leave.

The prompt is part of the design surface. On AI features I own the prompt together with backend: tone, context, and token cost. If a designer never reads the model's output by hand, the UI is just decoration.

Trust is built in steps. When you give a model real authority, you start people on a lower-risk version of it first. Skip that step and the trust never builds. And when AI acts for a person, feeling human beats feeling fast. I slowed a model down on purpose, so it felt less robotic and people trusted it more.

I review for how it feels. Reviewing my own work or a teammate's, I look at how it feels to use before I check whether the spec is met. No process gives you taste.