Making got cheap. Deciding got expensive.

I used to design in files. In this session I designed in a conversation that produced working software, and the difference turned out to be less about speed than about where my attention went.

Die Cut Studio is a parametric dieline tool for crafters: type a box’s dimensions, get a print-ready cut file at exact 1:1 scale. I already knew an agent could generate things. What I wanted to find out was narrower. If producing an option and changing an option both become close to free, what is left of my job?

So I took one sitting and worked on three parts of it at once: the editor, the brand, and the mobile experience. Everything below is from that session, dead ends included. The prototypes are embedded live, not pictured.

Prototyping in the real medium

The first question was where the 3D preview should live in the editor. That would once have been three static frames for me to imagine my way into. Instead I got three interactive mockups: real fonts, real colours, a box that actually rotates and folds.

I dragged the corner dock, it snapped shut into a tidy pill, and I knew in half a second. A comp describes an idea. A prototype argues for it.

Diverge wide, converge by reaction

Volume is the cheapest thing in this way of working, so I spent it freely. Converging looked different too. I was not writing specs. I was reacting.

Short, declarative reactions. The agent owned the mechanics: flexbox, media queries, React state. I owned which of them was right.

For the brand I asked for twenty wordmarks, each in a different open-source font, each working with the product’s own vocabulary: cut lines, kerfs, crop marks, and the cut, crease and bleed layers as a small dot triad.

Testing it where it has to live

I narrowed to the type-forward lockup, “Die Cut Studio” set in Bricolage Grotesque with the triad standing in for the “o”, then pressure-tested it in the two places that decide a mark: at nav size, and against different grounds.

The finalist was Blush: one light ground, one dark ground, the triad in the brand’s three layer colours, and a version on white for where it actually has to sit. Decided, then wired into the real Logo.tsx and the favicon in the same sitting.

Four layouts, one decision

The editor’s controls needed a new home. Four interactive layouts came back: a vertical accordion, an icon-tab inspector, dock cards, and free-floating cards over a full-bleed canvas.

I picked the floating cards, then refined them by reacting. The title moved into the toolbar, “collapse all” went away, the collapsed pills turned white, and a clipped shadow got fixed. None of those were things the agent proposed. They were things I noticed once I could touch it.

What the agent would not decide

It generates relentlessly, and it stopped every time a decision was hard to reverse. Solid rings or dashed. Tab bar or bottom sheet. How much blush in the UI. Each time it asked, with previews, instead of picking.

That division held all session. The agent produced variations, implemented them, changed them, tested them. What stayed with me was deciding what was worth exploring at all, setting the constraints up front, holding the options against each other, recognising the weak ones, judging hierarchy, and knowing when something was finished. When making gets cheap, those are the only moves left that cost anything.

Verification moved inside the loop

The agent tested its own work in a real browser and caught things a comp cannot show: pill shadows clipped by an overflow container; an Export panel that broke when collapsed, because an inline display:flex was beating the CSS meant to hide collapsed content; emoji icons that read inconsistently.

None of those are visible in a static file, because in a static file they do not exist yet. “Looks done” and “is done” stopped being separate states.

Mobile made the same point structurally. Floating cards cannot sit over the canvas on a phone, so the controls had to move to a FAB-driven surface. Three patterns to choose between.

Exploration and production shared one thread

The same conversation that explored all of the above ended with the chosen work in the live codebase: the new brand, the floating-card editor, a phone-first mobile editor with a Lucide tab bar and the full-sheet export preview. 273 tests still green, committed and pushed. No handoff, no translation loss.

Web

Mobile

What actually changed for me

The distance between designing and building got short enough that I stopped treating them as separate activities. That changed what I explore, because twenty options now cost what three used to. It changed how fast I iterate, because the thing I am reacting to is the thing itself. And it moved most of my attention off making the artifact and onto judging it.

It is worth being clear about what it is not. The agent over-produces, so curation is mine. It needs hard constraints up front: open-source fonts, IP-safe, the exact palette. It will confidently build the wrong scope if I do not confirm the scope. And it has no opinion about what is good. Range, fluency and tirelessness, but no taste.

I push fewer pixels and make more judgments. Every “this is the one” in this session was mine.

Making got cheap. Deciding got expensive. That is the whole change.