Vibe Coding Is Not Software Engineering: The AI Development Mirage
AI has made the atoms of code essentially free. But software engineering — the process — is as expensive as it has ever been.
The emergence of powerful AI coding assistants has introduced a new variation on the self-build temptation. The reasoning goes something like this: AI has democratised software development. Anyone can build an app now. We will hire a couple of graduates, give them Claude or Cursor licences, and build what we need in-house at a fraction of the cost. This is the citizen developer argument repackaged for the AI age. And it carries all the same flaws, plus some new ones. AI coding tools are genuinely impressive. They can accelerate development, reduce boilerplate, and help less experienced developers produce working code more quickly. But they do not solve the fundamental challenges of software engineering: architecture, design, testing, deployment, maintenance, security, scalability. The AI helps with the typing. It does not help with the thinking. There is a concept gaining traction among experienced software developers called the Faustian bargain. It starts as a prompt. The AI produces something that looks right, quickly. You feel powerful. But then you need to adjust something. Each fix introduces new edge cases. Each edge case requires another round of prompting. The complexity compounds. Before long, the productivity is through the roof, yet nothing really happens. And here is the part that firms forget: technology companies have access to the same AI tools. Every advantage AI coding assistants give you, they also give to professional development teams at software vendors. The delta has not changed. AI has made both sides faster. It has not made them equal. If anything, AI has widened the gap — professional teams know how to direct AI effectively, evaluate its output critically, and integrate contributions into maintainable systems. The gap between prototype and production has not changed. If anything, it has grown, because the ease of creating prototypes encourages people to underestimate the difficulty of everything that comes after.