Why I will not be using LLMs in my coding

Programmers are sharply divided over whether to use large language models (LLMs) for writing code: some report massive productivity gains, better test coverage, and the ability to ship more features, while others see buggier software, loss of craftsmanship, and a worrying erosion of problem‑solving skills. Many comments frame LLMs as a powerful but unreliable “junior developer” or power tool that still requires strong human judgment, especially for critical systems, maintainability, and long‑term code quality. Underneath the technical arguments lie broader concerns about job security, de‑skilling, ethics, environmental impact, and whether employers will eventually treat non‑use of AI as unprofessional or even irresponsible.

Adoption vs. Refusal

  • Many commenters see refusing LLMs for coding as equivalent to rejecting calculators or power tools: viable as a hobby, but increasingly incompatible with paid work.
  • Others defend abstaining as a legitimate personal choice, especially for preserving enjoyment of “hand” programming and cognitive engagement.
  • Several note that the stance may be sustainable only outside mainstream industry, where employers increasingly expect AI-assisted workflows.

Productivity, Quality, and Testing

  • Strong claims that LLMs massively increase throughput: features and refactors that once took weeks now done in days, tickets that sat for months now closed quickly.
  • Some report shipping large, AI-written changes without reading most of the code, relying on unit tests, benchmarks, monitoring, and user feedback as quality proxies.
  • Others push back: tests can’t cover unknown failure modes; performance or security regressions have been found only by reading the code.
  • Several argue LLMs are already better than many average or junior engineers at both writing and testing code; others call this exaggerated or “fake productivity.”

Effect on Skills, Roles, and Learning

  • A recurring fear: engineers who over-rely on agents will deskill and become “managers of interns that never learn.”
  • Seniors often liken LLMs to fast, tireless junior devs: useful if you already know where quality matters and how to review.
  • Concern that incentives to mentor juniors and build deep human expertise are fading; industry may converge on “adequate but sloppier” code.

Cognitive, Emotional, and Identity Issues

  • Some miss the satisfaction of solving problems and crafting elegant code; AI feels like a performance-enhancing drug that cheapens hard-won skills.
  • Others feel liberated: tedious boilerplate and side tools become feasible, freeing time for design, product, or non-work life.

Environmental, Ethical, and Corporate Concerns

  • Critics highlight energy use, data centers, and “skeezy” AI vendors; some argue ethics are unresolved today, even if solvable in theory.
  • Counterpoint: almost all modern tools have externalities; many companies are moving to local/open models for compliance and privacy.

Nuanced / Middle-Ground Positions

  • Several advocate a spectrum: use LLMs for boilerplate, tests, glue code, and exploration; write or closely review core, critical logic by hand.
  • Widespread agreement that code review and human judgment remain essential, even among heavy AI users.