Wix's $80 million vibe-coding bet just revealed its real fear: not that AI-built apps break, but that they all look the same.
You open ten apps built with the same AI coding tool and you notice it immediately. The beige backgrounds. The same rounded sans-serif type. The same button placement you swear you've seen on someone else's product last week. Base44 CEO Maor Shlomo told Business Insider that this sameness, not speed or cost, is what pushed his company to build its own large language model instead of just renting one from OpenAI or Anthropic.
"Everybody feels like they're getting the same UI when they're coding with the general models," Shlomo said. That's a blunt admission from the founder of a platform that exists to let anyone with no coding background generate a working app from a text prompt. Base44 lets you type what you want and watch it appear, database, authentication, deployment and all. If every app that comes out the other end looks like it was stamped from the same mold, the product stops feeling like creation and starts feeling like a vending machine.
UI and UX expert Paul Bakaus has a name for the look: an "algorithmic Uniqlo or Ikea." Not bad, not broken, just interchangeable. Bakaus told Business Insider the telltale markers are beige or tinted backgrounds paired with generic sans-serif typefaces, the visual equivalent of a hotel lobby designed by committee. Frontier models like GPT and Claude are trained to be good at nearly everything, from writing poetry to debugging Python, and that breadth comes at a cost. Ask a model to design a UI and it reaches for the safest, most statistically average answer it has, over and over, for every customer.
Shlomo's fix was Base1, a fine-tuned model built on an open-source base and trained on what the company describes as tens of millions of real user interactions from its own platform. "We think that if we take a model and we squeeze its ability to be really, really good at one use case, then we have a shot," he said. That's the whole thesis in one sentence. Narrow the model's job down to generating web and app interfaces, and it stops defaulting to the safest possible output.
Base44 didn't build Base1 in a vacuum. Wix bought the company for $80 million in June 2025, when Base44 was six months old and had all of eight employees, according to Wix's own announcement of the deal. Since then the platform's annual recurring revenue has gone from crossing $100 million to passing $150 million by May 2026, TechCrunch reported. That kind of growth curve draws competitors fast, and vibe coding has plenty: Lovable, which TechCrunch pegs at $500 million in ARR, along with Replit and Cursor, plus the frontier labs themselves. Anthropic's Claude Code and xAI's Grok can both write and ship application code now, which means the tools Base44 depends on could just as easily become the rivals that replace it.
Owning the model is Shlomo's answer to that squeeze. "Training and owning the model as part of our entire stack allows us a lot more optimizations on latency, cost, and efficiency," he told TechCrunch. That's the practical case. The design case is the more interesting one, because it's a bet that taste, not raw capability, is what customers will actually pay for once every platform can generate a working app in thirty seconds.
Base44 isn't walling users off from the big labs either. The platform still lets you choose Claude's Opus 4.8, Fable 5, or OpenAI's GPT-5.5 alongside Base1, according to Business Insider's reporting. That's a tell in itself. Shlomo isn't claiming Base1 beats frontier models at everything. He's claiming it beats them at the one thing his customers actually judge a product by on first glance: does it look like something a real designer made, or does it look like every other app built the same way last Tuesday.
"Models are progressing, but they'll stay very general in what they can do," Shlomo said. That's the bet other vibe-coding platforms will now have to answer. Lovable, Replit, and Cursor are all built on top of the same handful of frontier models, which means they're all exposed to the exact aesthetic homogenization problem Bakaus described. If Base44 is right that generic output becomes a liability once every founder is shipping AI-built products, owning a narrow, purpose-built model stops being an engineering flex and starts being the actual moat. Frankly, that's a much harder thing for a competitor to copy than a faster inference pipeline.
Also read: Nvidia Will Take a Cut of AI Startups' Future Revenue Instead of Cash Upfront • OpenAI Is Discussing Giving Washington a Stake Instead of Selling One • SoftBank pushes for a full $10 billion loan against its OpenAI stake
