Vibe Coding — Part 1: Behind the Vibe
AI coding has become so low-friction that a friend of mine literally sends prompts to Claude between Mario Kart races. And the crazy part? It works often enough to feel normal. That frictionless flow creates the illusion that complexity has evaporated. But once the “vibe” lands in the repository, dopamine ends and operational reality begins. And that’s where things get expensive.
What actually works:
Use vibe output only where failure is cheap. Prototypes? Yes. Core flows? Absolutely not.
Apply probability thinking. Ask: “How likely is the model to misunderstand what matters here?”
Recognize hidden rules. Concurrency, invariants, security boundaries — these are where vibe coding breaks.
Micro-insight: Speed isn’t dangerous. Uncontrolled speed is.
How much of your codebase today was written for momentum — and how much for longevity?