One system, three representations
GEL is Acquia’s cross-framework design system, built to unify several legacy systems into one platform. I’m its technical lead. The architecture starts with centralized design tokens, then carries them through shared Prime theming and framework-specific wrapper libraries built with PrimeNG, PrimeReact, and NX. Angular and React get native-feeling APIs without splitting the visual language in two.
That separation is deliberate. Tokens define what the system means, the Prime layer turns those decisions into primitives, and each wrapper presents the conventions of its own framework. A future framework can join without becoming the lowest common denominator for the ones already in use.
Scroll horizontally to inspect the diagram.
Read-only by design
The AI workflow around GEL treats tokens as an API contract. Engineers and agents cannot edit generated token files directly, which keeps Figma variables authoritative even when code is being produced quickly across many repositories.
Absolute rules fail when a release needs an exception, so GEL has a controlled override path. A contributor can temporarily adjust an individual token, while the tooling records the divergence and produces a report telling design exactly which source value should change and why. The escape hatch is visible, reversible, and useful to the team that owns the source of truth.
Scroll horizontally to inspect the diagram.
The path back to Figma
The Figma migration was agent-driven. Custom plugins handled the first phase, then Figma MCP let agents work directly on the canvas. That workflow converted a PrimeOne v3.1.0 fork that had diverged for a year into a wrapped PrimeOne v4 model. Code Connect now maps wrapped components back to their implementations, tightening the relationship between a design decision and the code that ships it.
Adoption without vigilance
Adoption is roadmapped across seven product surfaces, from Cloud Platform to Acquia AI. The wager underneath the architecture is straightforward. A design system scales when the easiest path follows the system by default. GEL is built to hold that line even when AI agents are writing a growing share of the code.