The potter and the machine

The potter and the machine

“A computer is an educational device, it’s in fact a direct reflection of your own imagination, your own intelligence. And once you’re given the freedom in which to create things and see the immediate response on the screen. Then it becomes a very enjoyable experience, you go on to involve yourself in many other things.”

— Unknown Computer Store Manager, 1979.

This quote played in my mind as I watched Cursor, an AI powered coding tool, execute on my vision for a Mac app to accelerate my team’s work. This project proved to be the perfect test as it had great potential without the risks of turning production code over to an agent, and the experience has fundamentally changed the way I think about what it means to design a digital experience.


For our design system, we define tokens, which are design decisions codified into variables which are shared across all platforms. They are essential for consistent application and efficient evolution of a product at our scale. For example, when it was reported that one of our components had inaccessible color contrast, we were able to push a fix simply by updating 1 token’s value. This required no additional effort from the various engineers who had implemented the component across all the platforms we support.

The friction, is in the creation of tokens. We use a popular platform called Style Dictionary for defining our tokens and publishing them in formats which can be consumed by all platforms. It’s a time consuming and error prone process to create these tokens because they are defined in massive JSON files. One misplaced comma or typo in a reference and your build fails. The errors aren’t very telling, so it’s very easy to lose significant chunks of time hunting for a small typo.

My idea was to wrap the editing of tokens in a GUI to eliminate errors. While I figured there was probably many JSON editing apps out there, I wanted to see if we could make something more particular to editing tokens and more closely aligned to our team’s particular processes.

My initial prompt didn’t give me a great app, but it was functional enough to prove that Cursor was capable of executing my idea. What ensued, was a brisk feedback loop between me and the GPT-5 based agent as my idea came to life before my eyes. Within the course of an hour, I had something pretty great. I still had a long list of ideas to make it better, but I was ready to put it to use.

“Seeing the immediate response on the screen” has always been the magic of computers for me. In the 70s, those bullish on computers lauded the technology’s ability to supercharge human ability, the bears heralded it’s potential to rob humans of the dignity of their work. This moment feels much the same. And while I’m greatly conflicted on where I stand from a moral and ethical standpoint, the reality is what it is, and my experience will always supersede any intellectualizing I can do on the subject.

That experience is bewildering. Magic, is the most appropriate word I can think of.

Rather than dissecting that experience, and what it may mean for the future of my craft, I’ll relate it to a hypothetical scenario:

Let’s say that instead of a digital product designer I’m an interior decorator, with 20+ years of experience. Over my long career, my process has changed quite a bit. I used to begin every design sketching on paper. I later produced higher fidelity mockups on a computer. Even with the computer as part of my workflow, I still sketch. I just find the process puts my mind and creative energies in a different place.

Then one day I’m given this magic power. I can walk into any room and speak changes into existence.

  • “Clear the room of all furniture”
  • “Apply X paint color to the wall”
  • “Fill this room with midcentury modern furniture”
  • “Let’s center align the sofa with fireplace”
  • “Swap the wooden legs of the sofa for stainless steel”

I’m convinced that if I had such an ability, and I was an interior decorator, I’d never sketch again.

Why would I?

A sketch, by nature, always fails to fully capture what it represents.

How could it?

Working in some mediums keep you close to the end experience. Sculpting with clay, you are right there in three dimensional space with the subject. Composing music for solo piano, on a piano, is likewise directly connected. So is typing a blog post.

Designing a digital experience is much closer than sketching a room. I’m looking at pixels on a screen just like users will. But how close do static pixels represent the experience? Only a flat, 2 dimensional version of it. This is why high-fidelity prototyping is so powerful. But it’s so expensive to do and I’ve found it’s hard to know where to stop. At some point, you’ve basically already made the investment required to actually build it.

Maybe that’s what this is, a new way to prototype… but it’s that air-tight feedback loop that makes this something altogether different for me. Like the potter reshaping the bowl with the slightest movement, and immediately feeling the result.

I think the future of design is having our hands in the clay, and once we do, we won’t find much meaning in sketching bowls.


By the way, I didn’t use AI in any way to write this post, and it probably shows. Writing isn’t my job, it’s more of an “art” for me. So I’ll leave in all the inefficiencies and idiosyncrasies. Thanks for reading!