Learn to code, ignore AI, then use AI to code even better

Original Article ↗ Hacker News Discussion ↗

Reaction to “don’t learn to code” / Replit CEO claims

  • Many see the claim that learning to code is a “waste of time” as marketing for AI companies and harmful messaging to beginners.
  • Several argue that, as with past hype cycles, engineers will still be needed; AI is another tool, not a replacement.
  • Some suggest a more accurate framing: learn to think and structure problems; coding is one of the best ways to build that skill.

AI as tutor vs. “vibe coding” trap

  • Strong support for using LLMs as always-available tutors: clarifying concepts, explaining snippets, walking through docs, and generating test data.
  • Multiple anecdotes (subreddits, a TikTok learner, college teaching) show beginners stuck with AI-generated code they don’t understand, unable to debug basic errors like non-existent methods.
  • “Vibe coding” (letting the model build everything) is widely described as a trap, especially for juniors: models write plausible but broken code and learners miss core mental models.

Effectiveness and limits of AI coding tools

  • Praised uses: boilerplate, unit tests, commit messages, merge request summaries, HTML/CSS layouts, low-level intrinsics, quick “remind me the syntax” answers, and small utilities.
  • Criticisms: hallucinated libraries/APIs, subtle bugs, unsafe refactors, confusion on framework idioms, wrong argument orders, verbosity, and context loss in longer sessions.
  • Some find free tools nearly useless and paid tools transformative; others see both as overhyped “slot machines”.

Skill, expertise, and what AI actually amplifies

  • Ongoing debate:
    • One camp: AI is a force multiplier for experts; you must be a subject-matter expert (in logic or language) or it will slow you down.
    • Another: it mostly raises the floor—great for low-skill tasks and non-coders who can write clear requirements, but unable to handle genuinely high-skill work.
  • Consensus that you still need the ability to specify problems clearly, reason about architectures, test, and debug; AI doesn’t remove the need to think precisely.

Career, education, and long-term concerns

  • Some fear AI will hollow out junior roles and create dependency on vendors; others note software jobs have historically expanded with productivity tools.
  • A professor describes adapting courses: no AI for basic exercises, structured use for larger projects, with emphasis on design, interfaces, testing, and systems knowledge.
  • Several warn that if few people truly learn to code, societies risk loss of technical sovereignty and stagnating training data and tools.