I’m genuinely curious about this and not trying to bash AI tools at all. As someone who manages a UX team, I’ve been exploring different AI solutions like Claude, Copilot, V0, Lovable and others to see if they could speed up our work.
What I keep finding is that these tools are great for creating new apps from scratch, but they don’t seem to understand existing design systems or component libraries (besides shadcn). They also don’t follow the specific UI patterns that already exist in established products.
Most of our daily work involves making improvements to features that already exist in a big, complicated software platform. We have to use components from our design system and stick to patterns that have been around for years (many aren’t even documented properly). The AI tools I’ve tested just can’t handle this kind of work in a meaningful way.
Has anyone found AI tools that actually help with improving existing products rather than building new ones? I’d really like to know what’s working for other teams in similar situations.
I’ve been using AI tools for about a year now, and you’re spot on about the biggest problem. These tools crush greenfield projects but fall apart with legacy systems and existing patterns. That said, I’ve had decent luck using AI for chunks of existing product work instead of trying to make it handle whole features. Like I’ll use Claude to polish copy variations when I’m updating flows, or brainstorm different approaches to IA problems within our constraints. It’s actually pretty solid at analyzing user feedback and suggesting small improvements to existing interfaces. The trick is dumping tons of context about our patterns and constraints upfront - basically creating custom prompt templates for our product. It’s definitely not seamless, but it helps with thinking through problems and cuts down exploration time. You still have to heavily adapt everything to fit your design system, but it throws out angles you might’ve missed.
Hit the same wall last year. We ended up building a wrapper around existing AI tools instead of fighting with them to understand our design system.
Made a knowledge base with screenshots and descriptions of all our components, then feed it to GPT when we need help with features. Takes setup time but now it suggests realistic improvements using our actual patterns.
AI’s been surprisingly good for research and analysis though. When improving features, I dump user interviews, analytics, and support tickets into Claude and ask it to find patterns or highlight focus areas. It catches stuff we miss in huge amounts of feedback.
For actual design work on existing products? AI tools still suck. But they’re decent for thinking through problems and organizing info. Don’t expect ready-to-use designs for complex legacy systems.
totally agree! AI tools are more hype than helpful in real-world setups. tried Figma AI too, it just doesn’t gel well with the nitty-gritty of established products. ChatGPT has been better for me tho! I lean on it for user stories and tidying docs while revamping features. not flashy, but it saves me time!