If AI is helping people code better, why aren't products getting better?

Hacker News Discussion ↗

Speed vs. Quality of Code

  • Many argue AI tools mostly make coding faster, not better.
  • Gains are largest for novices and in boilerplate, simple scripts, and glue code.
  • Experienced devs see “advanced autocomplete”: useful, but not transformative.
  • Some expect quality to decrease as less-skilled devs ship more code they don’t fully understand.

Code Quality vs. Product Quality

  • Strong consensus: better code ≠ better product.
  • Product quality is driven by UX, research, feature choices, and iteration cycles, not just implementation.
  • AI can produce “perfect code for a bad feature”; it doesn’t fix bad product decisions.

Who Benefits and How

  • AI is praised for:
    • Rapid prototypes/MVPs and throwaway apps.
    • Exploring unknown stacks, frameworks, and libraries.
    • Tedious transformations, config files, tests, and small automation scripts.
  • It behaves like a broad but shallow junior dev: decent at idioms, weak at deep, domain-specific problems.

Maintainability, Bugs, and Tech Debt

  • Concern that AI-generated code will be average, verbose, inconsistent, and harder to maintain.
  • Debugging vague, real-world bugs in complex systems remains hard; AI helps little there.
  • Fear that companies will replace senior devs with juniors + AI, increasing long‑term tech debt and bugginess.

Incentives, Enshittification, and Where Gains Go

  • Multiple comments: business incentives prioritize profit and feature throughput over quality.
  • Any productivity gains are often converted into “more features” or cost cuts, not polish.
  • Better tools historically lead to more software and complexity, not necessarily nicer products.

Timing, Adoption, and Unclear Effects

  • AI coding tools have only been widely used for ~1–2 years; many existing products predate them.
  • Some expect noticeable improvements in 3–5 years as new codebases started with AI mature.
  • Others see no clear evidence yet of large productivity or quality gains and suspect hype.

Overall Sentiment

  • Broad agreement: AI is already a useful coding aid and prototyping tool.
  • Disagreement on whether it “codes better,” and little belief that it has yet made mainstream products meaningfully better.