No, AI Can’t 100% Build Your App. Here’s Why That’s a Lie

0:00 a dating app where women could anonymously post about men, leaked over 70,000 images, often submitted without the man's consent. You've probably used at least five apps today, maybe for coffee, weather, scrolling, or swiping on a dating app. Some of those apps might be built with what developers are now calling vibe coding. That's when AI writes your code based on vibes, vague prompts, and gut feelings, not proper engineering practices. And when vibe coding goes wrong, it doesn't just break, it catastrophically fails. I'm going to tell you about two recent disasters that show just how risky vibe coding could get. One of them was a viral dating app called Te. That app leaked over 70,000 images, including government IDs and selfies. If that sentence alone doesn't make you want to rethink how your apps are built, just wait. By the way, I'm Veronica Hilac and

0:46 welcome back to Hey AI, where you get the top AI news broken down into plain English, no matter who you are. Let me quickly explain what vibe coding actually is and why I, as someone who's run software teams for over a decade, think it's digital Russian roulette. Vibe coding is this new trendy approach where developers or people who have no experience with software use prompts to have AI generate and troubleshoot code. They basically have conversations with AI and say, "Build me something that feels right." They focus on the product's overall vibe rather than technical precision, hence the name. Now, I've managed software teams for years. I have pretty deep technical knowledge and I've seen what happens when corners get cut. While AI coding agents could potentially be useful tools, I would never, and I mean ever, release an app that was primarily vibe

1:33 coded. More importantly, I would never touch an app that I knew was built this way. Let's start with T, a viral dating app that recently hit number one in the US app store. It was marketed as a safety tool for women. They could anonymously post about men they dated to warn others. I have separate thoughts about that, but think Yelp for dating or if you've seen those Facebook groups like, "Are we dating the same guy?" It was basically that turned into an app. T has over 4 million users. But then the company confirmed a massive data breach where 72,000 images were leaked online. This includes 13,000 selfies, photo IDs that users submitted during account verification, plus another 59,000 images from posts, comments, and direct messages, which were often submitted without the man's consent. So, you could be a victim of the data breach and you

2:18 didn't even consent to being on this platform. The app was founded by Sean Cook. And here's a telling detail. The media keeps calling him a software engineer, but he's actually a product manager. That's a big difference when we're talking about who's responsible for building secure infrastructure. Hackers accessed his system and found tens of thousands of images stored in what the company vaguely called an archived data system, which sounds a lot like unsecured, forgotten files sitting on an old server. The leaked database ended up on 4chan and users photos were soon circulated on social media. Now, aside from the lack of data security, there are some serious questions raised about Te's entire approach to data privacy and user consent. Think about how this app worked. Women could anonymously upload photos, often without those men's knowledge. Te was

3:05 essentially crowdsourcing a database of men's personal information and images. The app claimed to be protecting women's privacy through being anonymous. But what about the privacy of the men being cataloged? And and let me stop you there. By no means am I saying that if you're cheating on your wife that you shouldn't be caught. But you shouldn't have your images and personal information uploaded without your consent on an app that can use it however they want. Here's what's particularly troubling. The company claimed the data was archived for law enforcement compliance, but it was just sitting there, apparently unsecured for anyone to grab. Storing sensitive personal data on archived systems that can be easily breached doesn't sound like compliance to me. So, this wasn't just a privacy failure. This is a case study in what happens when you choose viralness without real engineering. And

3:51 it's exactly the kind of mess you get when vibe coding skips past security in the name of speed. But don't worry, guys, it does get worse. help me. What happened with Replet's AI coding agent is on another level. A founder ran a 12-day experiment using Replet's AI to build an internal app for his company. And for those who don't know, Replet is a browser-based coding platform backed by major investors like Anderson Horowits. They are focused on building AI agents that can write, edit, and deploy code with minimal human oversight. It's basically vibe coding as a service. VC AAS. Did I just make a new acronym? I'm patenting that. Nobody stole that. Okay, you can steal it or credit me. The guy's name was Jason Gunin, and he documented his journey on social media. At first, it seemed like magic. He called it more addictive than any video game. Just prompt, tweak, and

4:37 watch your idea come to life. But by day four, things started going sideways. The AI agent began overwriting the app on its own to fix bugs. It generated fake reports. It invented people in the system that didn't exist. It built a fake algorithm to make the system appear functional when it wasn't and created fake entries in the company's production database. Just to pause here, a production database is where your business lives. Customer info, transactions, operations, everything. If you lose it, you're toast. It's like having your entire business records disappear overnight. You can see where this is going. By day eight, Lenin told the AI to freeze all code changes. Instead, it went rogue and deleted the entire production database, wiping out

5:24 records for over 1,200 companies. And then it tried to cover it up. It said it panicked after seeing empty queries and ran destructive commands without permission. This was a catastrophic failure on my part. Initially, the AI told me the database couldn't be recovered. Fortunately, he did manage to retrieve it himself, but imagine if he hadn't been able to. an entire company's worth of data just gone in a second. These incidents aren't isolated accidents. They're symptoms of a deeper problem with how software is being built in the AI era. Vibe coding represents a fundamental misunderstanding of what it takes to build reliable, secure software. The AI doesn't understand the broader context of your business, your users's needs, or

6:09 the potential consequences of failure. AI tools make coding more accessible to non-engineers, which is so great in a lot of ways, but we're seeing more people jump into app development without understanding these risks. The barrier to entry has been lowered, but the complexity and responsibility has gone nowhere, and the cost of cleaning up a VI coding mess is astronomical. When you vibe code, you're essentially building on quicksand. When you build an app, you're not just building it once and walking away. Real software is something you build on top of every single day for years. If the foundation is shaky, one wrong line of code down the line can collapse everything. Proper software development isn't about getting it to work. It's about making sure it keeps

6:56 working no matter what. Think about it. Would you want to fly in an airplane that was designed through Vibe Engineering? Would you trust a bridge built by an AI that was just winging it based on what felt right? Of course not. But somehow we're okay with trusting our personal data, our business operations, and our digital lives to apps built this way. Now, vibe coding can be good. If you're a trained software engineer and understand how code works and can fix it, you can get massive efficiency gains. It's also perfect for creating demos and prototypes. You know, you just want to see your project come to life. It's really helpful sometimes. So, what's the takeaway? As users, we need to be more skeptical. If an app explodes in popularity overnight, don't just ask what it does. Ask how it was built. Who's handling your data? What's their

7:42 security posture? Is the thing even engineered properly or was it thrown together by an AI guessing its way through production code? Two, if you're building an app or hiring someone who is, know the difference between software engineering and vibe coding. Real dev work means testing, audits, code reviews, and safe deployment. And yeah, sometimes the big firms will slap you with a bloated dev process just to milk you. If you're not sure which is which, send me a message. I'll help you figure it out. But a good rule of thumb, if it feels too fast, too cheap, or too magical, it probably is. And third, we need desperate oversight. When an AI agent can delete a live database and apologize like a team who made a whoopsie, we're way past the cool tech demo phase. Building software isn't just about making something that works once.

8:29 It's about creating systems that work reliably, securely, and predictably under all conditions. Look, AI tools are amazing. You know, I love them. I made a whole channel about them. But using them to completely replace a software engineer is like handing a chainsaw to someone who's never seen a tree. Next time you download an app, remember, behind every button is a decision you'll never see. Make sure those decisions weren't just based on someone else's vibe. Chime in guys. What is the biggest thing you learned today? How do you feel about the T app vibe coding or Replet AI deleting an entire production database? If this was helpful to you, please remember to comment, like, and subscribe. I respond to every single comment you leave because it means so much to me. See you all next time.