Rebuilding and Operationalizing the Weedmaps Design System for AI
I led the transformation of Weedmaps' design system from a design-owned artifact into an engineering-enabled platform — establishing semantic token architecture, drift detection across ~201 tokens, a Figma-to-code pipeline, and CI/CD-backed governance, and partnering with Engineering to install a dedicated design systems engineering lead. Nothing about Weedmaps' visualized design had meaningfully changed in the prior 4-5 years; once this foundation was in place, we shipped the most meaningful design-system improvements in years within a span of weeks and months, and reduced design-system blockers to effectively zero.
The problem. When I joined, I found three different, disconnected views of the design system. Inside the design team, opinions were split — some considered it complete and essentially final; others had simply adapted to using it as-is, even though it had grown increasingly nuanced and complicated over time and wasn't as efficient as it could be. On the engineering side, most components were what I'd call "hand-rolled" — there was no central repository of official components, so engineers would copy whatever existing implementation they could find on any surface, assuming it was current. The result was product-wide misalignment with no real mechanism to keep design and code in sync. Nothing about Weedmaps' visualized design had meaningfully changed in four to five years.
My approach — treating it as a platform, not a library. Most teams stop at building a component library. I saw a larger opportunity: building infrastructure the whole organization could depend on. That played out in phases:
Foundation. I pushed the team toward semantic design tokens, component standardization, governance practices, and shared design/engineering ownership — positioning the system as a product, not a design artifact.
Engineering integration. I sponsored Storybook adoption and React Aria integration to close the gap between design and production code, starting with rebuilding our Button Primary component on React Aria foundations as the flagship example.
Token infrastructure. As Weedmaps expanded across products and platforms, manual consistency wasn't going to scale, so I invested in semantic token architecture and built drift detection across roughly 201 tokens to catch mismatches between design and implementation before they shipped.
Developer experience and CI/CD. Design systems usually become bottlenecks because every update requires manual review. I drove CI templates and automated validation so teams got the system's benefits automatically, rather than through centralized policing.
AI-augmented system. I recognized that AI tooling becomes dramatically more useful when it's grounded in structured design-system data rather than working ungoverned — so I made sure the system itself became a source of truth our AI tools (Claude, Figma Make, Builder IO Fusion) could reason against. This became the load-bearing prerequisite for our broader AI adoption work (see the AI adoption case study) — without it, AI-assisted design-to-code would have just produced more inconsistency, faster.
Platform strategy. I positioned the system as the stable foundation underneath dark mode, responsive web, and future AI-driven product work — moving it from "UI consistency" into a genuine organizational capability.
Making it durable. Recognizing this couldn't live in Figma alone, I worked with our VP of Engineering to identify and install a dedicated design systems engineering lead — a senior staff engineer and former creative director — who built the CI/CD pipelines and connected our components through MCP servers to our AI tools, starting with our most prominent UI elements (CTAs, product cards) and continuing today.
The outcome. We reduced design-system blockers to effectively zero. More tellingly: after four to five years with virtually no visible change to Weedmaps' design, we shipped the most meaningful design-system improvements in the company's recent history within a span of weeks and months once this foundation was in place, with designers shipping code not canvas representations.
Future state. In addition, we recognized the opportunity to create greater design cohesion by integrating the product design system into the brand system and the design of marketing teams. As our new Creative Director onboarded, we started with Claude Design and worked on the foundations of the alignment.