Case

OpenAI Codex harness experiment

February 2026 company article: a small team built and shipped an internal beta with no manually written code over about five months, reporting roughly a million lines and about 1,500 pull requests; early progress was slow because the environment was under-specified; the repository, not chat threads, became the system of record; the team enforced invariants mechanically. The company says results depend on this repo's tooling and may not generalise.

Industry
AI and software
Frameworks applied
Three waves: prompt, context, harness
Clipboard with a glass checklist panel, icon for consulting

Frameworks applied

This case is read through the frameworks below.

  • Three waves: prompt, context, harnessWhen a firm puts AI models to work, where should its leaders invest effort: writing prompts, engineering context, or building the harness around the model?

Sources

References

Listed A–Z by author surname, Harvard style. A source with a DOI links to https://doi.org/…; one without links to its page or says plainly that it has no DOI.

  1. Lopopolo, R. (2026) Harness engineering: leveraging Codex in an agent-first world. OpenAI.No DOIopenai.comSource page