In the past month alone I’ve heard: “I haven’t written code since 2025;” “Code reviews are dead;” “People no longer read code.” The industry is transforming, but people and organizations falling into the trap of no longer reading and writing code do so at their peril. Vibe-coded projects devolve into an unmaintainable mess.

Code maintainability and good architecture lack good measurements; it takes months, years even, to notice the effects. Bad code is hard to read, understand, and evolve. Changing one thing breaks the program in non-deterministic ways, resembling the “butterfly effect.” Adding a feature means changing code in multiple places and still forgetting to patch everything, creating inconsistencies. The design’s invariants aren’t clear, its authors are gone, and testing requires mocks, exposes implementation details, and prevents meaningful refactoring.

Experienced engineers detect code smells through intuition built with sweat and tears debugging production issues. Experts don’t follow rigid rules; they make the rules. AI isn’t trained on what maintainable code means. Reinforcement learning needs a reward signal measurable immediately, not in months or years, and most code in the wild is pretty bad.

LLMs are a great tool, but people relying on them to write and read code will never reach mastery. They no longer make choices, take responsibility for mistakes, or learn from them. Yikes!