OpenAI codes 1 million lines for an internal product
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
OpenAI built an internal product with over 1 million lines of AI-generated code using three engineers over five months. The approach relied on 'harness engineering' — structuring AI output through three techniques: context engineering (storing all knowledge as versioned markdown in the repo), architectural constraints (strict dependency layers enforced by linters to prevent module boundary violations), and entropy management (background agents continuously scanning and cleaning dead code). The team averaged 3.5 PRs per engineer per day across 1,500 merged PRs, claiming a 10x speed improvement over manual coding. The author notes the story reads more like a Codex marketing pitch than a verified engineering case study.