Evaluating LLM-based Agentic Approaches for Design Guidelines-Compliant User Interface Generation

Abstract

Recent advances in AI-powered user interface (UI) generation using large language model (LLM)-based agents enable users to create UIs from natural language descriptions. However, most existing systems struggle to satisfy enterprise-specific design guidelines, limiting real-world adoption. To investigate how LLM-based agents can more reliably generate design guidelines-compliant UIs, we compare three distinct approaches. The component-based approach achieves the highest compliance rate, outperforming the other methods despite moderately higher costs and latency. These findings suggest that generating UIs through reusable components is more effective for ensuring design guideline compliance than relying on LLM interpretation of textual specifications.

Related Projects

• Agentic AI 기반 UI/UX 생성-평가-개선 프로세스 자동화 기술 개발

  2026.09 — 2027.06