-2000 Lines of code (2004)

Original Article ↗ Hacker News Discussion ↗

AI-Generated “Slop” vs Crafted Code

  • Several comments link the story to current AI coding: Copilot/LLMs make it trivial to produce large volumes of “vibe-coded bloat” that technically works but is inefficient, over-abstracted, and hard to maintain.
  • People report cutting thousands of AI- or junior-written lines down to tens or hundreds, often with big performance and memory wins.
  • Concern that managers equate “more code written by AI” with productivity, mirroring the article’s faulty LOC metric.

Stratified Software & Quality vs Crap

  • Some envision a market split: cheap “hustle trash” software vs expensive, expert-crafted code (possibly with AI as a tool).
  • Others argue this already exists; the gap may just become more extreme, like artisan vs flat-pack furniture.
  • Debate on whether end users care about inefficiency (Electron, bloated apps): some say they feel it as sluggishness and slow bugfixes, even if they can’t name the cause.

Code Deletion as Real Productivity

  • Many anecdotes of large deletions: 8k→40 LOC refactors, 60k-line servers collapsed into libraries, hundreds of thousands of legacy lines removed via rewrites or consolidation.
  • Themes: code is liability/debt; best commits are often net-negative LOC; non-existent code doesn’t crash.
  • Some engineers pride themselves on being net-negative LOC over years.

Bad Metrics and Perverse Incentives

  • LOC, bug counts, “ticket touches,” and “% of code written by AI” are criticized as classic Goodhart’s-law traps.
  • Stories include bug-fix bounty schemes encouraging people to create bugs, and public “bugs caused/fixed” leaderboards that were successfully subverted.
  • Suggestions that any single-axis productivity metric (including “fewer LOC”) will be gamed.

Folklore Story Plausibility

  • Some doubt the literal details (“and then they never asked again”); others note the source is a direct participant and that high-status engineers often do get exceptions.
  • Consensus: whether embellished or not, the story captures a persistent truth about metrics that reward quantity of code instead of value.