Developers on Hacker News are sharing mixed experiences using LLMs like Claude for UI development, with many finding them excellent for CSS and basic layouts but struggling with complex interface design. The consensus is that LLMs excel at generating functional HTML/CSS and implementing specific design changes when given detailed instructions, but fall short when asked to create cohesive, modern interfaces from vague requirements.

Several developers report success using Claude with screenshot comparisons, feeding baseline designs from tools like Figma and iterating until the implementation matches. Others recommend using established component libraries like Tailwind and Radix UI rather than reinventing UI patterns. The discussion highlights that LLMs perform best when treated as coding assistants that need specific, detailed prompts rather than creative partners for abstract design work.

Some developers are building specialized tools to bridge this gap, creating AI-native browsers that integrate coding agents with browser automation for seamless UI iteration. The key takeaway is that LLM success in UI development depends heavily on the developer's ability to provide precise specifications and leverage existing design systems, rather than expecting AI to handle the entire design process autonomously.