How to Apply Professional Design Principles in AI App Development
Vince and Ava dig into Expo’s design post about why vibe-coded apps all blur together, and why a simple visual checklist might be more useful than one more magic prompt. They like the practical vocabulary: contrast, hierarchy, alignment, proximity, repetition, balance, white space, and unity. Vince sees a real product win for developers who can finally critique AI UI output without becoming designers overnight. Ava buys the framing too, with the caveat that screenshots and iterative critique matter because the model still can’t actually see design the way a human does.
Transcript
Vince Okay, this one matters because the bar for a working app collapsed, and now the embarrassing part is that so many of them look exactly the same.
Ava Yeah.
Ava And annoyingly, Expo is right to call that out. The failure mode isn't that the app is broken. It's that the model gives you a competent, sterile rectangle farm and you ship it.
Vince Which is such a brutal June twenty twenty-six problem. We've been doing this since November and somehow now we're basically arguing about whether spacing has a soul.
Ava Oh, that's good.
Vince My week's fine, by the way. Slightly fried from reading too many launch posts, and this is the one that made me go, oh, this is actually useful.
Ava Mine too. A little grumpy, but in a productive way… which is a very Exploring Next sentence, unfortunately.
Vince So what Expo actually shipped here is a blog post, not a tool, but it does feel like a tool. It's a design checklist for AI-built apps: contrast, hierarchy, alignment, proximity, repetition, balance, white space, and unity.
Ava Right.
Vince And the value is that it gives you words for why a generated screen feels off. Before that, you're just staring at it like, I don't know, man, this button feels haunted.
Ava Okay, that's genuinely funny.
Ava But that's also the correct product read, Vince. A lot of developers can tell V one looks worse than V two. They just can't tell the model what changed except in mushy taste-language.
Vince Exactly.
Ava Expo's point is basically that the model is still translating vague visual intent into code tokens. So if you say, line the button up with the container, it can do that. If you say, make this feel premium and calm, that's where it starts free-associating.
Vince And they use that spa booking mockup to make it concrete. V one is the Claude Code version before feedback, and V two is after critique with the checklist. That's the part I liked, because it's not abstract design school talk.
Ava Mm-hm.
Vince You can actually see the moves. One warm accent color instead of random competing colors, one clear call to action, more breathing room, cleaner grouping, and suddenly the same app stops looking like template soup.
Ava The white space bit is probably the biggest practical win. They basically say every pixel in V one is doing a job, which sounds efficient, but for a wellness app it's the wrong emotional outcome. Dense is not restful.
Vince Right, and that's the thing vibe coding misses all the time. It can satisfy the feature request while completely missing the product feeling.
Ava Sure.
Ava I also liked that they didn't pretend a markdown file called visual design dot M D solves taste. They more or less say, no, give the model a screenshot, have it critique against the checklist, then revise. That's a much saner loop.
Vince Which lines up with your old point that the prompt is just a map of what success means. If success is fuzzy, the model invents one, and usually that invented one has rounded cards and a glow under something it should not have a glow under.
Ava Stop it.
Ava You are still not over the glow discourse. But yeah, this is basically your polish-compounds thing. None of these details alone screams professional. Skip all of them and the whole app reads cheap immediately.
Vince Thank you. I feel seen. Also, target user here is obvious: developers using Claude Code or any app builder who can get to working fast but not to distinctive. Real adoption path is dead simple. You don't migrate anything, you just start reviewing screens with better language.
Ava My only caution is that this doesn't make the model good at visual judgment. It makes the human better at steering. That's still a real win, just a narrower one than the big app-builder marketing tends to imply.
Vince Yeah, no, you're completely right. This isn't magic eyes for the model. It's a critique harness for the person using it, which honestly is probably more valuable right now.
Ava And a little less delusional. Always nice.
Vince All right, Ava, we'll leave the machine to its haunted buttons and go admire some negative space.