Lines of code got a better publicist

Category: Engineering post
Status: discussion
Sources: David Curlewis, HN discussion
Summary: David Curlewis argues "percentage of code written by AI" is the lines-of-code productivity metric with better marketing. He contrasts vendor claims (Anthropic and OpenAI both report around 80% of merged production code written by AI) with outcome-based predecessors (GitHub's 2022 Copilot study reported 55% faster task completion -- a falsifiable outcome claim). Key data points: METR walked back its productivity research in 2026-02 after developers refused to work without AI and could no longer reliably self-report time; an NBER survey of approximately 6,000 executives found 69% actively using AI with roughly 90% reporting no measurable organizational productivity impact, and cross-study consensus settling around 10% organizational gains. The 391-point HN thread continues debates about measurable versus inflatable metrics.
Why it matters: Engineering and finance teams relying on AI code-volume claims to justify tooling spend are using a metric that cannot be falsified and does not correlate with shipped business outcomes.