Ask HN: How do you employ LLMs for UI development?

Hacker News lit up over whether large language models (the chatbots powering code helpers) actually help build user interfaces. The original poster says AI made their site changeword.org look better and even ran a screenshot feedback loop to fix pixels until they matched. Others doubled down with tools like Claude’s frontend-design skill, Figma add‑ons, and Chrome helpers to auto-compare screenshots. Cue the drama: one camp loves AI for the “boring bits” — scaffolding menus, converting specs, polishing CSS — while the other says it can’t design its way out of a paper bag. The hottest take: if you can draw the final look, “turning it into CSS is trivial,” which sent designers into orbit. Skeptics warned the pretty code hides landmines: confident styling that quietly breaks interactive parts, sneaky security mistakes, and code that ignores your project’s structure. Meanwhile, tool-heads flexed with Playwriter to let agents poke at the browser, and workflows that compare screenshots as a pass/fail test. Taste vs. tooling became the meme of the thread — a few said “don’t invent new patterns,” stick to Tailwind and components; others joked AI has vibes like a very confident intern. One delightful metaphor: build the frame like a house, then let the bot paint — but don’t let it pick the furniture.