AI-Assisted Design System Workflow
From user stories to design-system-driven interfaces
I built a Proof of Concept to test whether an AI-assisted workflow could translate product requirements into editable Figma interfaces while preserving an existing enterprise Design System as the source of truth.
I curated pairs of previous user stories and their approved Figma outcomes as few-shot / in-context examples, giving the model concrete references for how requirements had historically translated into UX patterns, components and layouts. The workflow combined that contextual guidance with explicit UX rules and structured Design System information — components, variants, variables, tokens and interaction patterns — reaching the system through an MCP-based context layer instead of inventing UI from scratch.
What I tested
Running a real user story through the pipeline, the agent produced editable Figma views built from existing Design System components — not generic UI. The translation held: component reuse and UX intent survived, and the output was reviewable and editable rather than a flat mockup.
Status
Validated POC. Not taken to production — the goal was to prove the translation layer worked, not to automate design decisions. The designer retains validation and final judgment.
- Domain
- Design System Automation
- Length
- POC · 2026
- Role
- Design & Execution