I don't start my designs in Figma anymore

Figma is open on my other monitor right now. I'm in it most days. I just don't start my designs there anymore. Nowadays, I start in the terminal, with a prompt and a cursor blinking back at me like it has notes.

A few years ago, building something you could actually click took a few days. That's too expensive for something you might have to throw out. Now it takes a few hours. Instead of one polished prototype, I'll make a few rough ones that I can quickly click through before I commit to a direction.

Motion beats stillness

The clearest case was a globe I built for Forma. It takes every kilometer you've run and wraps it around the Earth, so your training shows up as one long trip, city by city. You can grab it, you can spin it, you can watch the line crawl across the surface.

I knew exactly how it should look. I didn't know how people would use it: which way they'd turn it, where they'd grab it, how fast it should spin. You can't answer that from a still.

Once it was running, the surprises started. I kept reaching for the globe itself to turn it, not the controls. So the whole thing became the handle, and it opens on your own city now. That held up when I handed it to a few other people.

The trail surprised me the same way. It draws itself onto the sphere as the distance adds up, something I could never have built by hand in Figma. Down to how it wraps around the back. So I fed it sandbox distances and let the awkward cases sort themselves out.

Trying it on a real screen

The globe was one isolated component. The real test was a fuller screen, built from a dozen smaller parts. Forma runs on an engine that turns every workout into three numbers: fitness, fatigue, and form.

The fitness chart is how you read it: three lines, one for each, with a row of daily-load bars underneath. It tells you whether you're fresh or running on empty. It's the screen I open the most.

I built the first real version in an evening. A few years ago that would have eaten most of a sprint. But the bigger shift came after it shipped. I reshaped it three or four times over the next few days. Each pass was a quick prompt and a fresh build, in the time a single Figma comp would have taken. I think the screen got better mostly because changing it stayed cheap.

The loop I settled into

I describe the thing, run it, and click around until it shows me something I didn't expect. Then I throw it away and write a sharper version. What I keep is what each one taught me: which edge case actually mattered, which interaction felt right, which idea looked good on paper and fell apart in my hands.

The first version that comes back is almost never right. It reaches for the obvious layout, the one I'd have drawn myself, and quietly skips half the states I asked for. So I see where it breaks, write down what's missing, and ask again.

I'm still learning to write the sharp version. The loose one comes first by default: plain language, thrown down fast, because it's quicker and because that's how the thought actually shows up. The precision comes after. Spelling out the roles, the states, and the things to avoid only becomes possible once the intent has settled in my head, and by then the prompt is mostly just my plan, made explicit.

Where Figma sits now

None of this retired Figma. It moved later, to the work it does best. It's still faster for breadth. I can lay a few directions side by side and take them in at a glance. A prototype makes me click through them one at a time. And once an interaction is settled, Figma is where I turn it into a system, the components and tokens that carry one decision across forty screens.

Lately the two have started feeding each other. I can pull a prototype's screens into Figma through its MCP, work on them there, then hand them back to the model to build from again. Prototype, Figma, model, and round again. It stopped being a one-way handoff a while ago.

Escaping rabbit holes

The fitness chart is the one that got me. Each pass was cheap, so every new question felt worth entertaining: a color for every sport in the load bars, whether a run and a swim on the same day should stack into one bar or blend into one, whether clicking a bar should open the workout itself. None of it was the question I came in with.

I know I've gone too far when I'm fixing something nobody will ever see. A glitch in a state that can't happen. A console warning. So I try to stop the moment the prototype has shown me what I needed. Sinking into the build feels like progress, and I fall for it more often than I'd like.

In some ways all of this has made the work better. In others I've swapped one set of problems for another, and I'm still working out which is which.