My first professional work was in graphic design and illustration. I learned to code because I could not afford to hire developers to turn that work into websites, and the technical path grew from there. This project is a return to that starting point: using visual design to shape the software rather than treating it as a finishing layer.
The goal
I am building a reusable design system for several enterprise applications in one monorepo. Each application should remain independent while sharing a consistent visual language, interaction model, and set of components.
The exercise is also a practical comparison of current design and coding tools. I want to understand how well they carry intent from an early visual concept into production code, and where human judgment still has to intervene.
Figma and Stitch
Figma is capable, but connecting its design system to coding agents took more setup than expected. Its generated concepts were useful as references, though they often looked complete before they were structurally reusable.
Stitch offered a faster conversational workflow and a simpler connection to the implementation process. For this project, it was the more fluid tool. The tradeoff is consistency: generated visual work can drift sharply between sessions, even when given strong references.
What I learned
The useful workflow is not “generate a finished interface.” It is to identify repeated regions, turn them into explicit components, define where variation belongs, and test whether the system can support more than one screen.
AI can translate intent, suggest patterns, and expose inconsistencies. It cannot decide what is worth keeping. That remains the designer's role, especially when generated interfaces converge on the same rounded, polished, vaguely familiar aesthetic.
The outcome I want is not a collection of screens that look finished. It is a system that can be understood, extended, and recomposed without losing its character.