Building a Design System Bridge with AI
This project started as a VP-level request to create static whitelabel components in Figma to justify further investment in a design system.
I realized pretty quickly that static artifacts wouldn’t answer leadership’s real questions. The risk wasn’t visual quality. It was scalability, maintenance cost, and whether this system would actually be used.
The real problem was that our design system lived across multiple Figma files with no single source of truth. Small changes were expensive, documentation drifted, and engineers still had to translate intent manually.
So instead of just shipping components, I designed an AI-enabled operating model that lets designers validate real components directly in code. Figma became a collaborative exploration tool, and code became the validation layer.
One key decision was enforcing a component demo-first workflow. Designers had to push a real component, interact with all its states, and resolve edge cases before any documentation was written. This reduced rework and improved AI output quality by keeping context focused.
Another decision was introducing structured session planning and command-based guardrails. Non-technical designers weren’t allowed to just prompt the AI freely. They had to agree on scope first, which implicitly taught systems thinking and prevented chaos.
To prove this could scale, I let a junior designer with no coding background own a real component end to end. I intentionally stepped back. He planned, built, debugged, documented, and shipped independently. That success signaled to leadership that this model could extend beyond designers to product owners and developers.
The result was compressing idea-to-validation time from months to days, shipping multiple production-ready components, and establishing a living design system as the source of truth. What this unlocked for me is a new way of working. I now focus on designing operating models where AI removes friction between disciplines and makes validation fast and cheap at scale.