Ask HN: Do you know of any company that went back to hand-written code?

Hacker News Discussion ↗

AI coding tools are praised for rapid prototyping and reduced “inertia,” but many engineers report that the speed gains are offset by exploding technical debt, unstable “legacy” codebases formed in months, and loss of deep understanding of systems. Commenters describe divergent strategies: some teams ban or severely limit AI for core code while using it for reviews, tests, or boilerplate; others are pushed by management to maximize AI use even at the cost of quality and long‑term maintainability. A recurring theme is that the real issue is less the tools themselves and more leadership, incentives, and workflows that prioritize short‑term output over code quality, learning, and sustainable engineering practices.

Scope of the Question

  • OP asks: are there companies that adopted AI code generation and then reverted to human-written code?
  • Several replies say they know of none at scale; others note many firms haven’t adopted AI coding tools at all yet.

Productivity Gains vs Code Quality

  • Some argue AI clearly boosts short-term productivity by removing inertia and speeding prototyping.
  • Critics say “productivity” is meaningless if it just produces low‑value or wrong features.
  • Many predict long‑term “gains” will evaporate under extreme technical and cognitive debt.

Tech Debt, Legacy Code, and “Slop”

  • Multiple comments describe AI as a “nuclear-powered footgun”: you can build a legacy codebase in months instead of years.
  • One startup story: AI-enabled rapid iteration led to messy, unstable core code; a large refactor failed because AI kept reintroducing old patterns via git context. Team is considering banning AI for core parts.
  • Others report teams losing project knowledge when 99% of changes are AI-generated; hard bugs take much longer.

Limited or No-AI Policies

  • Some teams avoid AI for core/“deep” code and only use it as a reviewer (akin to static analysis).
  • A startup with ~15 engineers hand-writes all “interesting” core logic, using AI only for commodity UI and search-like tasks.
  • Another company uses no AI at all; others only allow AI for security/performance checks or tests (with skepticism about AI-generated tests’ value).

Leadership, Process, and Culture

  • Several argue messy AI-driven code is a leadership/process failure, not an AI inevitability.
  • Suggestions: enforce structured workflows for AI use, emphasize communication and shared understanding.
  • Tension noted between “happy team using AI how they like” vs long-term software quality.

Comparisons, Analogies, and Other Domains

  • Some liken AI tools to IDEs, compilers, and Stack Overflow; others counter that AI “hallucinates,” so the analogy is weak.
  • Report of doctors abandoning AI scribes: any time saved was lost verifying verbose, inaccurate notes.
  • Concern that widespread AI reliance erodes learning, craftsmanship, and overall product quality, even as corporations optimize for short-term output.