“As code throughput increased, our bottleneck became human QA capacity.”

— Ryan Lopopolo, OpenAI · Increasing application legibility

Quote · openai.com

From “Harness engineering: leveraging Codex in an agent-first world”, found 2026-10-14.

Take

The fix was to make the app bootable per worktree, with an ephemeral logs, metrics and traces stack, so Codex could run and check one instance per change itself.