A couple of years ago, I'd have drawn my process as a loop — one stage after another, all the way round to assess, then start over. That loop is gone.
Now it's become a mesh: research, design, building and testing happen almost at once, feeding each other in real time. And because I build as well as design, I can take an idea all the way to something live on my own.
That speed is the risky part. When output is this fast and this parallel, it's easy to ship something that only holds together on the surface. So the grip matters more, not less: the user is the fixed point, and keeping vision and usability steady is what keeps the whole thing honest.
I direct the tools hard and trust them little. They make the output; they don't make the calls — the design, the usability logic, the data behind every number stay mine. I'm honest about my scope: I direct the build, I don't write the code. But directing it well is its own craft — knowing what to ask for, catching what the model gets confidently wrong, and stopping before the polish outruns the thinking.
I still love the old craft, and I keep my hands in it — the tools are there to sharpen the work, not to do the thinking. Used with judgment, they don't replace the craft — they raise the ceiling. I used to be fast. Now I'm faster, and I can do more.