Ask HN: Why Is AI Coding Hype So Different From Developers’ Experience?

Summary

A Hacker News post questions the growing claims that AI is already better than humans at programming, that programmers are becoming unnecessary, and that anyone can build software through “vibe coding.” The author says software produced this way has often been poor in personal experience, while supporters describe major productivity gains. One commenter contrasts software with electronics hardware, where detailed documentation and costly compilation encourage more rigorous design; they argue that LLMs benefit from access to library documentation and discussions of observed bugs, helping them produce individual routines quickly and accurately. The same commenter says humans are still generally better at connecting those routines into a reliable architecture. Other participants warn that a generated application may compile and appear functional while containing redundant code, logic errors, and race conditions that a user cannot see. Some argue that AI is better than most people who attempt coding, while others say its speed changes the economics of development but does not eliminate the need for competent review. The discussion also notes that tools such as Claude and Codex can sometimes locate bugs quickly with careful prompting, but may behave unreliably during debugging. The original poster says AI is useful for reviewing snippets, yet letting it operate freely on a codebase produced poor results.