Don't Bother with Vibe Coding

Original Article ↗ Hacker News Discussion ↗

Definition and Perception of “Vibe Coding”

  • Thread notes that “vibe coding” originally meant letting an LLM write nearly all the code, “accept all” changes, barely reading diffs, and poking it with error messages until things run.
  • Several commenters are confused by the term and distinguish it from normal AI-assisted coding (autocomplete, rubber-ducking, targeted snippets).
  • Some feel the article and critics are overreacting to a meme or buzzword; others think using it seriously in job posts or academia makes those institutions look unserious.

Reactions to the YC “Vibe Coder” Job Ad

  • The Domu “Vibe Coder / AI Engineer” listing gets heavy criticism: long (12–15 hour, including weekends) days, low/now-raised pay, onboarding via making debt-collection calls, and “50%+ of your code written by AI” as a hard requirement.
  • Many initially thought it was parody; others see it as VC/hype theater rather than a serious engineering role.
  • The product focus—automated debt-collection voice calls—is widely described as dystopian and “vomit inducing,” with worries about harassment and even suicides.
  • The post is taken as emblematic of exploitative, bro-ish startup culture and of YC backing “shitty people,” though some think pile-ons are unproductive.

Where Vibe Coding Might Fit

  • Broad agreement that “pure” vibe coding is acceptable for: prototypes, weekend projects, solo demos, small internal tools, and low-stakes glue code.
  • Several note that for these, AI assistants are genuinely “life-changing,” drastically lowering activation energy.

AI-Generated Code Quality and Limits

  • One camp claims they can have tools like Cursor/Gemini/Claude generate large amounts of SOLID, tested, production-ready code indistinguishable from their prior work, with humans mainly reviewing and prompting.
  • Skeptics doubt this, arguing that fully AI-driven codebases are bloated, miss non-obvious constraints (security, performance, compliance), and break down on maintenance and bugfixing.
  • Many say LLMs often fail at “fix this existing complex code” tasks; maintenance is where real engineering value lies.

Professional Engineering vs. Hype and Career Pressure

  • Commenters stress that “coding” is the easy part; the hard parts are infra, CI/CD, ownership, testing, documentation, long-term support—areas where LLMs can help but don’t replace judgment.
  • Some argue seniors must master AI-assisted workflows or fall behind; others counter that no single skill is that urgent and warn against panic and hype-chasing.
  • Overall sentiment: AI tools are real and powerful, but “vibe coding” as “AI takes the wheel” is risky beyond small, disposable projects.