Making a Design System AI-readable
We worked with a globally recognized brand company to scale their design across multimodal experiences with AI
- Role
- Lead Designer, research
- Year
- 2026
Problem: Design Systems were traditionally created to be read by human designers to provide rules and constraints for making branded things. This process is well documented and understood- with books like Atomic Design and Refactoring UI being written about the topic. Today, Design Systems are more often becoming read by Large Language AI Models (LLM) and the translation doesn't quite work as easily as giving the LLM a design system. Design Systems need a new way to exist to be used by LLM consistently- or else, there is risk of serious technical debt and churn.
Approach: We performed user research on people who are trying to use LLM to design things within an enterprise setting. Tools like Claude Design and Figma make provided a competitive baseline to understand. For our primary research, we asked participants to step their experience of creating with AI. We transformed those findings into a strategy that the team could use to improve their design system. We launched a Jobs-to-be-Done interviews and then a prioritization survey to understand what to prioritize.
Result: Findings and insights were produced and recorded in a research report. These findings and recommendations were scaled to senior members of the organization who used that information to make strategic recommendations and operationalize priorities. A lot of serious work has been done to make the experience of using the Design System with AI better, specifically through the org's website and the resources it provides to builders- improving productivity by 20%.
Lessons: The lines between Designers and Engineers is blurred more than ever. It's easier than ever to transform an idea into a real working digital thing. This comes with its own risks. The rules and consistency of Design Systems are compromised when LLM veer off of the course of the Design System. This comes with serious operational costs and technical debt, and potentially higher maintenance costs- which is why making these design systems LLM-readable is so important.