Code as the Medium
With AI, design's medium is code. Here's how I'm thinking about and acting on the opportunities.
The old medium
For two decades, digital product design's "medium" was design software that rendered design work on a computer screen in a proprietary way. It was a series of artifacts (informed by research) defining the product: wireframes in Axure, designs in Sketch, then both in Figma; prototypes stitched together from artboards and hotspots; spec files handed off to engineers.
From time to time a question would pop up: should designers learn to code? The answer that stuck was practical. Learn enough to collaborate well with developers: the basics of operating systems, frameworks, and languages; a feel for architecture, and for where the front end stops and the back end starts; a sense of what you get for free versus what has to be custom-built.
Design and development were each a full-time discipline, each with its own training and depth. Anyone genuinely excellent at both was rare, and expensive to hire. So the split held, and no one was breaking the door down to build an open-source Figma clone that outputted code from designs.
One pattern I've watched over a decade of design leadership (first at an agency, now in enterprise) is what happens when disciplines share a medium. Every time designers, engineers, and product managers moved into the same tool, collaboration went up. Cross-learning went up. Empathy went up. The shared medium afforded new interdisciplinary insights that were previously very difficult to access.
The tool is never the magic — everyone finally working in the same medium is.
The tool is never the magic — everyone finally working in the same medium is.
What changed
AI collapses the distance between "the old way" of designing toward something we're all hot to call "building."
I can sit down with Claude Code and a Next.js scaffold and have a running prototype by the end of an afternoon. With the right amount of preparation and the correct support structure, I could likely do this faster than I could if I started from scratch in Figma. I have been putting this technology through its paces: the FEMA events table in my Compliance Agent case study is code. ScreenYeet, my screenshot-to-action app, is code.
Two months of commits, zero traditional workflows in Figma. I've also been excited to experiment with "AI-native" design systems and design tools. The work I'm sharing lately lives at a URL, and that has me genuinely excited.
204 contributions in the last year
The bigger shift: more people can become builders. We don't know exactly what that means yet. Right now I'm most excited for the benefits of the shared medium, and for the focus of the work moving from "is the visual ready?" to does it work for the end user?
The best part isn't that I can build faster — it's that more people can build at all.
The best part isn't that I can build faster — it's that more people can build at all.
How I'm taking advantage of the shared medium
When filling in visual and interaction details gets cheap and fast (especially with a mature design system doing some of the heavy lifting), designers get time back for the hard work.
The hard work I'm having my team do right now is strategic: choosing what to build and what not to build. The user-centric research needed to make that call, and the training to deploy design in service of this kind of strategy. The other hard work worth mentioning is building the ecosystem. As we build AI-powered solutions, we're building with AI tools — and designing the process and tooling ecosystem around ourselves to optimize for human-centered outputs.
Code as the medium means my team can spend more time on strategy, research, early design, and usability testing. We're getting feedback from users earlier and more often. We're running studies against software that actually exists. We're using real and realistic data. We can stop waiting for someone else to show us how our designs will really work.
In code, you confront the real shape of things: how all the disparate parts fit together into a coherent whole in the way we intended.
What's also changing is the volume and complexity of what we're building. Agentic tools don't mean less software. They mean more of it, and more of it is adaptive, stateful, and alive. A system like that resists being fully described in a static file, and even a well-built prototype only holds still long enough to snap a photo of it. More and more, the fastest way to specify a living agentic system is to build it.
Let's connect.
Open to conversations, collaborations, and meetups.