Has anyone figured out how to use large language models (LLMs) for actual frontend development? I'm not referring to creating shadcn-based UIs or vague prompts like "design me a website." I'm talking about using LLMs to replicate precise, existing designs, ensuring accurate spacing, typography, and layout—no loose approximations allowed. Extra points if LLMs can adhere to a specific coding style and respect an ongoing codebase with established patterns. Has anyone found success with this, and what setups or workflows worked for you?
3 Answers
In my opinion, aiming for absolute precision in frontend designs is unrealistic. Designs are tailored for specific screen sizes, but browsers vary widely. Expecting perfect spacing and typography across the board doesn't hold up when faced with real-world usage scenarios.
Honestly, after using LLMs for a while, I don’t think they can implement designs with extreme precision just yet. They still need a lot of guidance from us to get it right, like step-by-step help.
I've had some good results using LLMs to write tests and generate repetitive code based on templates I set up. It helps take care of some of the mundane parts!
That’s not accurate. Absolute precision and exact spacing do follow set standards, like 1rem being consistent across devices. Good design takes all screen sizes into account from the start. But to answer your question, LLMs still aren't capable of this level of implementation at the moment.