Build Log 1: I’m kicking off my next big feature for LoFi: the component system. While looking for inspiration, it feels like Figma’s system is still the only game in town. But to be honest, I’ve never liked how Figma handles components. Here’s why: 1. It makes the user work too hard Take the humble Button component, probably one of the most standardised elements in any UI. Sure, the style and behaviour might vary slightly by platform, but 90% of the time it’s the same. So why make me build it from scratch? You could argue Figma’s system is flexible for customisation, and sure, it is. But software isn’t plastic. You can offer full flexibility and ship components like buttons and forms with smart defaults and customisable options out of the box. 2. Too many dropdowns and toggles The way we customise components is through toggles, dropdowns, nested panels, feels like a solved problem that no one’s bothered to rethink. What if components were context-aware and adapted to the environment they’re placed in? This is a hard problem to solve. I’ll need to get back to you on that one. 3. Figma components are built for UI/Dev, not for design thinking Figma’s components focus on structure,size, states. But they ignore design intent. They don’t help you answer: What makes a button convert better? What tone or language should I use here? When should I use urgency? How should I pair buttons to guide decisions? None of that is baked into the system. But it could be. --- The Challenge So yeah, the goal is to build a better system than Figma’s. No pressure. I’ve given myself 2-3 weeks to ship it. Can I pull off something creative? Honestly, I don’t know. This shit looks hard. If you’ve got any ideas, throw them my way. I’m starting from zero here. lol.

To start the build, I wanted to try something different: getting different AI models to pitch me solutions. I copied and pasted this post for context and told the models I’d go with whoever gave me the best answer. I tried three models: Grok 4: This model understood the pitch best and worked the hardest. But it only expanded on my thinking. It didn’t bring any new or great ideas to the table. Sonnet 4: It didn’t try very hard. It just built on my thoughts and suggested practical ways to implement what I already had. Nothing new here either. OpenAI o3: This one didn’t even try. Total nothing burger. I ended up going with a recommendation from Grok 4, take a premade system (Shadcn), parse it into LoFi’s JSON format, and build on top of that. I gave Grok 4 the job. I did a pretty bad job parsing the components though, so I scrapped the build and started over with another model. I tried Sonnet 4 again. It failed too. I’ve concluded that these AI models are struggling because they aren’t familiar with LoFi’s layout system. They don’t know what 'good' looks like, so they can’t automate the process properly. I’ll have to manually create a few solid examples first, then try automation again later. I need a new solution.

Demo of Lofi - a FREE wireframing tool. I’d love to hear your feedback! 🙂 No sign-up required, and your data is saved locally on your computer. Link in comment