I jumped in hard and gave away way too much control.
I'm talking about designing and building in this new AI-supercharged era. I started in a very "can I build this?" place: spun up a rough version, gave the system light direction, and let it make a lot of the decisions. Layouts, flows, the general shape of the app.
The speed is useful, but it can make you give away the very decisions that make a product feel considered.
It was exhilarating. The speed was intense. I was building, actually making, and suddenly had an app I could play with.
The app worked. It just didn't feel like mine.
But the trade-off was that it all just looked like iOS.
Which is fair. It was using those components and those patterns. Still, that was the first reckoning for me. Speed had got me to something functional, but it hadn't got me to something that felt like mine.
So I pivoted. I started bringing more design into it: handing over components, setting base styles, piecing together the beginnings of a system. My thinking was that I could shape the foundations and let the system handle the rest.
After all, that is what we're gaining, right? Speed, and an intelligence that can help craft and shape the product while we focus on the important bits.
That worked somewhat.
But after a bunch of iterations, I realised I was the bottleneck. More specifically, my design time was the bottleneck.
At first that felt like a problem. Then I realised I was looking at it wrong.
This is where I should be spending the time. If the building part is becoming almost mindless and easy, the design part still isn't. Not if I want the work to live up to the level of craft I care about.
If I want quality, I have to lean on the knowledge and experience I already have. This is my job. I know how to design a good app. I do it every day at work.
I spend time animating the loader. I document the screen states. I build the component system. I make the calls on what should feel quiet, what should feel obvious, and what needs another pass.
AI can help with some of that, but it can't do it to the level I want by itself.
So that's what I've done. I've stepped back. I still use Claude Code for contextual work because it understands the main structure of the app, but we're not building the app right now. We're in the design phase.
That is what I'm doing now. The app exists. Now I need to make it feel like it was designed.
You know, maybe this is what people mean when they say AI looks most capable outside your own discipline. It can feel like it knows the work, until you know enough about the work to see what is missing.