In OpenAI’s Holiday Extras case study, the team reports producing a working app rebuild in two weeks using Codex with engineers, then moving from annual major releases to weekly releases in the first month. Open-to-book conversion increased 32.5%, while the error rate fell from 19% to 9%. Engineers retained responsibility for architecture, APIs, and security, supported by small documented reusable packages and parallel iOS and Android development. These figures compare different periods, so they do not isolate Codex’s causal effect or establish a universal productivity multiplier.