You open the pull request. It’s 5,000 lines. It’s massive. It’s a masterpiece of modern AI generation. And it’s completely, utterly broken because the guy who wrote it never bothered to run it.
AI didn’t break your codebase. It just made your bad engineers faster.
We’ve all been there. A partner, a colleague, a junior dev gets high on the hype of “vibe coding”—that intoxicating feeling of talking to an LLM, watching the text stream in, and feeling like a 10x developer. They hit enter, the AI spits out a PR, and they immediately push it to production. They feel like a genius. You feel a migraine coming on.
The debate raging across tech right now is whether AI can actually code. That’s the wrong question. The real crisis isn’t about the machine’s capability; it’s about the human’s competence. Vibe coding isn’t the problem. The problem is treating production as a publish action instead of a verification process.
Friction used to be the safety net. AI removed the friction, and now the floor is falling out.
Before AI, writing code was hard. It required thought, structure, and iteration. If you didn’t know what you were doing, the compiler would scream at you. You had to test because the effort to write it demanded you verify it. Now, a prompt generates a solution that *looks* right. It has perfect syntax. It has confident variable names. But it’s a hallucination wrapped in a syntax error.
If you are shipping untested, vibe-coded PRs to production, you aren’t a visionary. You are a liability. You are taking your own incompetence, wrapping it in an AI bow, and making it someone else’s emergency.
Speed without guardrails isn’t innovation. It’s just technical debt accumulating at the speed of light.
The partner in this scenario isn’t a bad developer because he used AI. He’s a bad developer because he skipped the fundamental step of software engineering: verification. If he wasn’t going to test his code with AI, he wasn’t going to test it without AI. The LLM just let him mask his laziness behind a wall of volume.
We need to stop romanticizing the vibe. Good engineering isn’t about how fast you can generate code; it’s about how rigorously you can prove it works. If you want to vibe code, go play in a sandbox. If you want to build software, test your damn code.
FAQ
Q: Isn't AI just a tool? Doesn't the fault lie with the developer?
A: Exactly. The tool isn't the problem; the process is. AI just scales whatever engineering culture you already have. If your process lacks verification, AI makes the failure faster and bigger.
Q: What's the practical takeaway for teams using AI?
A: You must enforce a hard line between generation and production. AI can write the code, but human verification, testing, and code review must be non-negotiable. If it's untested, it doesn't ship.
Q: Is vibe coding ever actually useful?
A: Yes, for prototyping, brainstorming, and internal tooling where breakage is acceptable. The moment it touches a production environment without tests, it stops being vibe coding and becomes malpractice.