# Baemin's CLAYDesign Aims to Replace Three Screen Redraws With One

> Woowahan's design system team describes how a general-purpose AI model generates screens from its design system, targeting the plan-design-code handoff.

- Publication: Thoai Dispatch
- Canonical URL: https://vokythoai.dev/article/baemin-claydesign-ai-design-system-screen-generation
- Section: News
- Author: Woowahan / Baemin Tech (KR)
- Published: 2026-11-20
- Tags: design-systems, ai-code-generation, baemin, woowahan, design-to-code, frontend, product-development

## What happened

Woowahan Brothers (Baemin) published a post on its engineering blog describing CLAYDesign, a project from its design system team that uses a general-purpose AI model to generate screens from the company's design system. The post, titled "세 번 그리던 화면을 한 번에: CLAYDesign이 디자인시스템으로 화면을 생성하는 방법" (roughly, "One screen instead of three: how CLAYDesign generates screens with the design system"), appeared on the [Woowahan Tech Blog](https://techblog.woowahan.com/28538/) on November 20, 2026.

The article frames the problem in terms of repetition. Before a single screen ships, a planner, a designer, and a developer each draw it once, so the same screen is drawn three times. When planning changes during development, the wireframe, the visual comp, and the code all have to be revised again. Because each stage redraws from the previous stage's decisions, the finished screen can drift from the original plan, either because intent is miscommunicated or because implementation diverges during development.

## Why it matters

The three-draw workflow is a familiar source of rework in product organizations, and the post treats it as a design system problem rather than only a process problem. If screens can be generated from the same design system components that the final product uses, the argument goes, the gap between the plan, the design, and the code narrows, and changes made during development do not automatically trigger a full redraw across all three artifacts.

This is also a notable framing choice: rather than presenting AI as a replacement for designers or developers, the post positions a general-purpose AI model as a consumer of the design system. The design system supplies the vocabulary, and the model assembles screens from it. That places the approach closer to constrained generation than to open-ended image synthesis, though the post excerpt available here does not detail the generation mechanism.

## Technical details

The source material available for this summary is limited to the post's introduction and framing. It states that the design system team used a general-purpose AI model, but the excerpt does not name the model, describe the architecture, or report benchmarks, latency figures, or adoption metrics. It also does not specify which Baemin services or screen types CLAYDesign has been applied to, or whether it is in production.

What the material does establish is the scope of the problem the team set out to address: three parallel representations of one screen (wireframe, design comp, code), repeated revision when requirements change mid-development, and intent loss across handoffs. The post's title indicates that the design system is the mechanism by which the AI generates screens, which implies the system's components and rules are inputs to generation rather than outputs of it.

Readers who want the implementation specifics, including the model choice and any evaluation of generated screens, will need to consult the original post at [techblog.woowahan.com/28538](https://techblog.woowahan.com/28538/).

## What to watch

Three questions are worth tracking as this work develops. First, fidelity: whether generated screens match the design system's constraints closely enough to be used as a shared reference by planners, designers, and developers, rather than as a starting draft that still requires three redraws. Second, change propagation: whether a mid-development planning change can be applied once and reflected consistently across the artifacts, which is the specific failure mode the post identifies. Third, scope: whether CLAYDesign remains an internal design system team project or extends to product teams across Baemin's services.

The post is a statement of approach rather than a results report. Until the team publishes model details, evaluation criteria, or usage data, the claim that three draws become one should be read as the project's goal, not a measured outcome.

## Sources

- [Woowahan / Baemin Tech (KR)](https://techblog.woowahan.com/28538/) (2026-11-20)
