Everyone's scrambling to learn the next AI tool. But I'm increasingly convinced the biggest thing holding back AI-assisted design isn't your prompt skills — it's your design system.
Let me explain.
What happened
There's a growing conversation right now about making design systems AI-ready, and the central argument is deceptively simple: the quality of AI-generated design output is directly proportional to the quality of your design system.
Well-structured tokens, clearly named components, and thorough documentation don't just help human designers — they give AI tools the context they need to generate outputs that actually match your intentions. Without that structure, you get drift. The AI produces something that looks roughly right but is subtly wrong in ways that take longer to fix than doing it from scratch.
The focus is on three practical goals: reducing the gap between what you intended and what the AI produces, minimising errors in generated prototypes, and maintaining consistent context across AI-assisted workflows.
None of this is revolutionary. But that's exactly the point.
Your design system is your most important prompt
We've been framing AI readiness as a skills problem. Learn to prompt. Master the tools. Keep up with the latest model. And sure, those things matter.
But I think it's actually an infrastructure problem. Your design system is the prompt — or at least, it's the most important one. If your tokens are inconsistent, your component naming is a mess, and your documentation reads like it was written in 2019 and never updated, AI tools will faithfully amplify all of that chaos.
Think about it this way. When you hand a new designer your design system, how long does it take them to produce work that's consistent with the rest of the team? If the answer is weeks, AI tools are going to struggle just as much. Possibly more, because they can't pop over to someone's desk and ask what "card-alt-v2-final" actually means.
The teams who'll benefit most from AI-assisted design aren't necessarily the ones with the best prompt engineers. They're the ones with the cleanest systems.
This connects to a broader shift I've been watching. The most impactful AI work in design isn't happening at the generation stage — it's happening at the systems level. Building structures that are machine-readable as well as human-readable. Encoding decisions into the system itself rather than relying on tribal knowledge that lives in someone's head.
We've always said good design systems should be self-documenting. Now there's a very concrete reason to actually follow through on that.
Tool spotlight: CSS contrast-color()
Here's a small but powerful example of what self-correcting design systems look like in practice — and it's not even an AI feature.