You can’t imagine it if you notice your websites starting to converge into one beige sans-serif haze.
One of the biggest benefits of AI is that non-technical people can vibecode their ideas into real, monetizable apps. As I wrote in April, anyone can build an app in hours using tools like Claude Code, Lovable, Replit, and Base44.
These AI-designed apps have some tell-tale signs, and the devil is in the details. They use similar design styles that look pretty but are dysfunctional. Your app may work on a small scale, but these small details can become big issues when you scale up and commercialize it.
Here’s how to tell if your app is AI-coded and how to change it.
1. Regression to the mean, aka the painful middle
First sign: The app’s design is boring and unconventional.
Paul Bakaus, CEO of AI design startup Impeccable, said in a June 23 podcast interview with venture capital firm Andreessen Horowitz that the perks of AI, especially for Claude Design, include beige or tinted backgrounds and sans-serif fonts.
He calls it “the Uniqlo or Ikea of algorithms,” and while it’s not a bad design, it’s not necessarily unique.
Donghoon Shin, a human-computer interaction researcher at the University of Washington, published a paper on how vibe coding led to the homogenization of design.
Singh told Business Insider that mood-coded products tend to converge to a “single aesthetic that is statistically average.”
Features: A muted color palette with lots of whites and grays, a single branded accent color, standard sans-serif typography, and elements with rounded corners and drop shadows.
When we built a test app using Base44 to act as a photo editor for a newsroom, the design was heavy on beige elements and sans-serif fonts. Screenshot/Aditi Varade
Sauvik Das, associate professor at Carnegie Mellon University’s Human-Computer Interaction Institute, told Business Insider that this is a “reversion to the mean” effect.
Those trying to make money with AI-encoded apps are also aware of this.
Priyanshi Bansal, a product manager in India, used Claude to vibecode an app that helps people choose gifts for their loved ones. early Users told her the app looked like “AI slop.”
She said the first version of her app included a lot of emojis, shadows and rounded edges.
“I’m currently building version 2 of the app, which is much better, especially the UI. I’m a product designer, so I need to get this finished,” she said.
2. Cute and dysfunctional
Screenshot of an AI-coded subscription tracker in minutes. BI
Second sign: The website is pretty fancy for a product that isn’t yet finished.
“For example, you’ll see a very polished landing page for a product that’s still in early alpha stages,” Das says.
It may look pretty, but many features don’t work intuitively because AI is designed with an emphasis on aesthetics and less on ease of use.
Ankush Samant, a lecturer in digital innovation and design practice at the National University of Singapore, said user interface and experience designers are trained to read human emotions, behaviors and intentions. They know exactly how to design “the weight of the buttons, the pace of the onboarding flow, the tone of the error messages.”
“AI tools are optimized for a happy path and tend to produce interfaces that look complete until someone actually uses them,” Samant says.
Das says that vibe-coded products are That part has no function. For example, when you hover your mouse over an element, the element’s outline becomes subtly shaded and increases in size indicating that you can click to see more information, but when you click, nothing appears.
3. Error 404: Ignoring special cases
And finally, Vibe Coding Tools spares no consideration for edge cases.
“Designers also spend a significant amount of time on what I call edge-state design: empty states, error messages, skeleton loaders, and offline states. AI often postpones these or skips them altogether,” Shin says. Said.
Samant said how an app presents errors can tell whether the app’s creator has thought through the user experience or whether it’s still in the demo stage.
“AI tools tend to either skip these altogether or generate placeholder copy that says ‘Something went wrong. Please try again,’ removing the human voice precisely when users need reassurance the most,” he said.
Indeed, the vibecoding startup notes all these shortcomings.
San Francisco-based startup Base44 launched its proprietary AI model, Base 1, on Monday. We hope this reduces the erratic appearance of the AI in vibe-coded products and produces a better UI/UX than the Frontier model.
I vibecoded the app. Well, what is it?
In this house, we will not embarrass the vibe coder. We are big enthusiasts ourselves.
However, if you find that your app has some of the above red flags, here are some ways to change them.
Samant said the first step is to stop looking for beauty and start driving decisions.
“Instead of ‘I want this to look clean and modern,’ think, ‘This screen is where users who are concerned about their data decide whether to proceed. What do I need to delete? What do I need to copy?'” he said.
Second, atmosphere coding is not all about atmosphere. Singh said users need to provide the AI tool with as many details as possible, including design references, brand constraints, and what users don’t want to see.
Also, if you’re looking to scale up, you may need to hire experts to take your product to the next level.
Das said that while AI is a useful tool for improving user interfaces, it may not be the best choice for creating great UIs from scratch.
“That’s not to say that a purely vibe-coded app can’t have sustained commercial success, because the most important thing here is product-market fit,” he said. “But I think you always need a good UI/UX designer to make a good product great.”