I talked to a data scientist the other day. He was proud—genuinely proud—of his productivity. He was paying $200 a month for Claude Max, and he’d used it to build dozens of apps. Almost fifty, he said. Some were even live. Then I asked him one question: “Where’s your GitHub?”
Silence. Then a confession: He didn’t know how to use Git. Or GitHub. He had nothing published, nothing versioned, nothing that anyone else could look at. He’d been spending all day “making projects” with AI agents, but he had never committed a single line of code to a repository.
AI isn’t democratizing software engineering. It’s creating a class of ‘prompt-ware’ builders who can generate massive output without any understanding of the fundamental discipline that makes software sustainable.
You’ve probably noticed this too. The flood of AI-generated apps that look impressive in a demo but collapse the moment you try to add a feature, fix a bug, or collaborate with another developer. The people behind them are often spending hundreds of dollars a month on AI tools, building at a furious pace, but they’re skipping the boring, essential parts: version control, code review, testing, documentation.
This isn’t a story about one data scientist. It’s a pattern. The AI coding boom is producing a generation of developers who can generate code but not understand it. Who can build from scratch but never maintain. Who can create but never collaborate.
And here’s the twist: The emperor has no clothes, and he’s paying $200 a month for the privilege of not knowing it. The very productivity that AI tools unlock is masking a deep, fundamental incompetence in the basics of software engineering. The higher the output, the less incentive to learn the discipline.
But let’s be clear: this isn’t the fault of AI tools. It’s the fault of a mindset that mistakes output for value. The data scientist wasn’t a bad person. He was simply taking the path of least resistance, and AI made that path incredibly smooth. The problem is that smooth paths don’t build good engineers.
For the rest of us—the ones who actually read code, who use Git, who review pull requests, who understand that software is more than a prompt—the message is clear: Stop measuring yourself by how many apps you can build. Start measuring by how many you can maintain.
The future belongs to engineers who embrace the friction, not the ones who avoid it. Because when the AI bubble corrects, the ones who can’t tell you what their code actually does will be the first to disappear.
FAQ
Q: What's wrong with building apps without knowing Git?
A: You're building on sand. No version history means you can't roll back, can't collaborate, can't track changes. It's the difference between a prototype and a professional product.
Q: Is this really a widespread problem?
A: Yes, especially among non-engineers and junior developers who rely entirely on AI tools. They produce volume but lack the foundational skills to maintain or scale what they build.
Q: What should I do differently if I use AI coding tools?
A: Learn the fundamentals. Use Git. Read the code the AI generates. Write tests. Don't let the tool replace your understanding—let it augment it. The goal is to be a better engineer, not a faster prompter.