February 12, 2026
Lines, Lies, and Earnings Calls
Lines of Code Are Back (and It's Worse Than Before)
Big Tech brags about “more code”; devs ask, “more bugs”
TLDR: CEOs are boasting about how much code their AI writes, reviving the discredited “lines of code” metric. The community fires back: it’s bloated code and bad incentives, with a split between nuance (LoC for project complexity) and pragmatists who say success is happy customers, learning teams, and working software.
The internet lit up after a piece warned that “lines of code” — the infamous bad metric — is back thanks to AI. CEOs are flexing percentages of AI-written code like gym selfies: Google says 30%, Microsoft says 20–30%, Meta’s gunning for 50%, and Anthropic claims up to 90%. The crowd’s reaction? Equal parts side-eye and popcorn. One running joke: we’re measuring progress by “code weight” again. Another: “steps app for developers,” counting lines like daily steps.
Commenters went in hard on quality. “AI cranks out bloat” was the vibe, with users noting it’s easy to inflate numbers while bugs and rewrites don’t make the slides. The Goodhart’s Law chorus returned — when you target a number, you break the number — and a viral tweet bragging “100k lines to build a C compiler in 2 weeks” became a meme after a Community Note pointed out reality. Beginners like kittikitti admitted they once clung to line counts for validation, only to feel mocked later — a raw, human subplot. Nuance sneaked in too: some argued LoC can still estimate project complexity, just not people’s worth. But the knockout comment came from the outcomes crowd: don’t count lines, count “happy clients, growing teams, and profit.” The only metric that shipped was the clapback.
Key Points
- •The article asserts that lines of code (LOC) was long considered a poor productivity metric and recounts historical criticisms from prominent figures.
- •It reports that tech CEOs now publicly cite the percentage of AI-generated code at their companies (Google, Microsoft, Meta, Anthropic).
- •Tooling like GitHub Copilot and Cursor emphasizes line-based metrics, and the article claims 256 billion lines of AI-written code were generated in 2024.
- •Social media examples highlight LOC counts as achievements, including claims of a 100k-line AI-built C compiler and individual outputs measured in millions of lines.
- •The piece frames the resurgence of LOC focus through Goodhart’s Law, warning that targeting code volume can undermine meaningful measures like quality or bug rates.