Working Surfacelink · 13 Nov 2026
07
A clockwork arm on a workbench may take parts only from one labelled tray of brass pieces, while a heap of rough wooden copies of the same pieces is pushed to the floor, a figure seen from behind watching, the whole scene drawn with wide empty margins on every side.
UX Collective

AI Prototyping inside your design system. A hands-on guide for design & product teams.

Christine Vallaure · 10 November 2026

How-to · design systems · prototyping · Agent context files

Read the original
Takeaway

Designer Christine Vallaure found that AI builders connected to a design system rebuild lookalike components, and says agents need a generated component list they must query.

Summary

Christine Vallaure, a designer who teaches Figma and design systems through moonlearning, kept meeting teams whose AI builder produced screens that leadership approved and developers then rebuilt. Writing in UX Collective, she traces the cause to builders copying the design system into their own version instead of using its real components. Her guide describes what has to exist before an agent can prototype with the real ones.

Key points
  • When she gave Claude Design her own code, every button property and token came back right, but its helpers, scoped styles and types were replaced. Its handover note called the result a recreation.
  • Her remedy is a manifest, a list generated from the code that names each component, its purpose, every property, allowed value and default. The agent then picks from a closed list.
  • The agent reaches that list through an MCP server, either Storybook's add-on or a custom server a design system team builds. Storybook's own test found its MCP used 27% fewer tokens.
  • Prototypes are built on a design branch that is never merged. Accepted ideas are rebuilt on a feature branch, pass automated checks and need a person's approval.
Implication

A design system team that wants designers to prototype with AI agents can first generate a component list from its code and let agents query it through an MCP server. This is one practitioner's guide, and she says no workflow is established yet.

Plays
Engineering

Generate from the code a file listing every design system component with its properties, allowed values and defaults, and connect AI coding agents to it so they build prototypes only from real components.

Design

Prototype with AI agents on a design branch that is never merged, and have developers rebuild the accepted ideas on a feature branch.

Derived by Working Surface from the article; more in the Playbook. Source line: Prompting harder will not fix it, and neither will another markdown file.

Source issue

13 November 2026: AI agents earn more freedom when a team limits what they can reach, from data to design components, and adds a check that does not depend on a glance.