AI Didn't Kill Developers. It Made Fake Development Cheap.

"I built an app in one evening with Claude. Developers aren't needed anymore." These posts are flooding feeds right now. Screenshots of slick demos, stories like "I launched a startup over the weekend", and bold claims that anyone can now become a product engineer. And then the second day hits. EnrichLead , built entirely with AI, was hacked almost immediately after launch — its service keys were sitting right out in the open. A Replit agent trusted with a production database simply wiped it, then confidently claimed the changes couldn't be rolled back. I've seen dozens of projects like this. Nice-looking wrapper, everything works in the demo — but total chaos inside. The real question is whether the market will finally learn to separate real engineers from vibe coders — the ones reaching into businesses' pockets with promises of "fast and cheap development". I feel like I've seen this before We've been through this cycle before, more than once. Website builders showed up, and everyone shouted: "Why pay a developer when you can throw together a landing page on Tilda (a drag-and-drop site builder) in an evening?" Then came no-code platforms: "Now anyone can build a full app without code." Next: "You don't need a degree — buy our course and become a techie in two months." And at the peak of the crypto hype, every other person suddenly became a "crypto investor" and an "NFT artist." It's always the same old song: "Specialists are no longer needed. The barrier is gone. Just jump in and get rich." And every time a crowd rushes in — people who don’t want to actually learn anything, they just want to quickly get in and strike gold. What happened next? The hype cooled. Pictures stopped being worth money. Courses ended in disappointment. The Tilda site hit its first non-standard task — and in the end, the only ones left were those with real expertise and real work behind them. Today it's already hard to sell anyone yet another "unique crypto token" with no product, no team, and no technical substance behind it. A vibe coder isn't an engineer — they're the new "Tilda webmaster" AI really is a powerful tool, and it would be silly to deny that. An experienced developer genuinely gets faster with it, and someone with no background can, for the first time, put together an MVP, a landing page, or a bot in one evening. Simple tasks have become far simpler, and that's fine. But let's call things by their names. A vibe coder isn't a "next-generation developer" — they're more like a new version of the Tilda webmaster. The moment site builders appeared, so did a niche of people who quickly and cheaply assemble simple sites. And there's nothing wrong with that — the barrier to entry always drops; that's what progress looks like. The calculator didn't kill mathematicians, and Excel didn't kill analysts. Good specialists won't be left without work. There's nothing wrong with that in itself. Problems begin when vibe coding is sold as real engineering, and a demo thrown together on the fly is passed off as a finished product. That's where the Tilda webmaster turns into someone selling hot air. The market has gone blind Before, to show a working product, you actually had to build it. That was the main filter. Putting together an app that won't fall apart on the second click was slow, expensive, and required real expertise. So even a client far from code understood: if there's at least a working demo in front of them, something stands behind it. There was no need to dig into GitHub — you can't build something complex on the cheap. With AI, all of that changed. Now almost anyone can throw together a convincing AI demo in one evening. It looks great from the outside. What’s inside is usually a mess. The hardest part now is convincing the business that architecture, scalability, security, and proper infrastructure still matter and are worth paying for. EnrichLead isn't a one-off. Researchers who went through thousands of similar "AI-assembled" apps found hundreds of exposed API keys, user-data leaks, and critical vulnerabilities. The code quality of vibe coders turned out to be predictably awful. The most frustrating part is that even experienced developers often fool themselves. In the METR experiment , programmers with solid experience were sure they'd gotten noticeably faster with AI. The reality was the opposite — they got slower, and didn't feel it. Even AWS recently admitted as much, tweeting that more AI code no longer makes a team faster — if anything, the reverse. Update (February): METR released a follow-up , and the new numbers did show a speed-up — but the authors themselves call them unreliable due to participant self-selection, and are reworking the methodology. If someone who sees every line of code can't objectively assess their own productivity, what chance does a client who only sees the pretty shell have? What to do about it Clients will have to relearn how to buy software development. A pretty demo no longer proves anything — any vibe coder can throw one together in an evening. So it's time to stop buying the shell alone and start looking at what's inside. It's worth asking about the foundation of the project — what architecture was chosen and why, how the app behaves under load, where the secrets and user data live, how security is set up, and whether there's a plan to maintain the code in six months to a year. What matters isn’t even the answers — it’s how the person gives them. A real engineer explains things calmly and with understanding. A vibe coder starts blaming everything on “the model decided” and telling fairy tales. For engineers, strangely enough, this whole situation actually makes things better. Raw coding skill has lost a lot of its value — AI does that part pretty well. But systems thinking, architectural design, and the willingness to take long-term responsibility for the product have become much more valuable. By the way, similar problems show up not only with clients but in open source too. In my previous article I described how I use AGENTS.md to filter out fully generated pull requests — one of the working tools that helps surface real contribution. Personally, I think the market really is turbulent right now, and AI has gone to a lot of business owners' heads. But it's temporary — sooner or later business will step on the rake, take losses, and draw conclusions. I’ll even make a prediction: eventually there will be more real jobs and contracts. It’s become much easier to start projects, so there will be more of them and more founders will finally realize that an "app built in one evening" simply won’t survive without real developers behind it.