LLM Agents for Design System-Compliant User Interface Generation: A Comparative Study of Context Engineering and User Prompting Strategies

Abstract

Large language model (LLM) agents are increasingly used for user interface (UI) generation, yet enterprise settings require compliance with organizational design systems. However, it remains unclear how design system knowledge might be provided to agents, and how practitioners actually interact with them. Study 1 compares three context engineering strategies—instruction-based, context-based, and registry-based—for design system-compliant UI generation. Results show that the registry-based strategy achieved the highest compliance and visual similarity, while other strategies were less accurate but more cost-efficient. Study 2 identifies three prompting strategy types—Full-spec Starter, Layout-first Builder, and Low-spec Starter—of practitioners by characterizing real-world interactions with a deployed LLM agent-based design system-compliant UI generation, revealing differences in upfront specification and subsequent behavior. Together, these studies offer empirical guidance for selecting context engineering strategies based on desired compliance granularity and for understanding how users interact with LLM agents for UI generation.

Related Projects

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

  2026.09 — 2027.06