Why I Built a Design System for AI
A specific system built to adapt.
Most design systems don’t cover AI well enough. States like agent orchestration, execution failures, streaming responses, agentic diagrams. Some systems handle a few of these, but not all of them, and the ones that try tend to stumble over themselves when different executable states conflict in hierarchy.
One thing I admire about Anthropic is the simplicity of the terminal window they’ve designated for Claude Code on a personal machine. The simple use of color, the simple use of bolding or animation. I wanted to emulate this experience in a design system. Keeping things simple, nearly binary. Plenty of room to create or modify, but simple enough to understand everything at a glance.
The Problem
This simplicity was something I’ve craved when working with IBM’s Carbon design system. Google Material design solved some of these problems with its flexibility but still didn’t quite feel correct at all turns.
I’d been using minimalist UI cards in Figma for years on enterprise engagements to inform my decisions when creating UI or navigating UX issues. It worked, but a lot of questions were answered by my own experience, and not everyone has the expertise or prior knowledge to come to the same conclusions.
That being said, a design system doesn’t need to answer every question, in fact, it shouldn’t. It’s a framework. Too many times I see mid level designers looking to a design system like a paint by numbers guide, when in reality it’s more like driving a car down a country road. We are all given a list of rules to follow but there is some leeway when the road signs start disappearing or there is no longer a painted line on the asphalt.
Where Does AI Come In?
AI is similar right now. The UX hasn’t been strictly determined yet. There’s talk about Generative UI, adaptable interfaces that change as the user adopts certain habits or favoritisms. Generative UI is a cool concept but I think right now it fundamentally goes against human behavior. In a few years maybe that’s where we will be at, but the human in the loop can’t make determinate decisions if the answers are a moving target.
What we do now as designers is embrace change and fall back into our design principles and our design systems. A new process or user behavior floats to the top? Make a change or map out that UX, add it to your library. Nothing new here. What is new is designers feeling that they shouldn’t change. That’s where the true issue creeping into our design systems. I’m not saying get a LLM subscription. I’m saying the job has always been solving user problems, and that doesn’t stop because the medium shifted. If anything, it’s the reason to lean in.
Charcoal Design System. A work in progress design system