Escalate to a human only when judgment is required
OpenAIPublished by the organization itself (post), no named author
Post: Harness engineering: leveraging Codex in an agent-first world, OpenAI (own site), February 11, 2026.
Filed under
OpenAI sells AI or agent tooling, so it has a reason to publish flattering numbers about AI. The passage stays in the archive; read it with that in mind.
How this was checked
Checked against the source by a first reader, then confirmed independently by a second reader who fetched the source without seeing the first reader's notes (Independent review agent (Claude Sonnet), blind second pass from passage text and citation only, no repository access, 2026-10-31).
Wayback copy, because the live page gave 403. 'Escalate to a human only when judgment is required' is an item in a list of agent behaviours and matches exactly. It is OpenAI's own post.
How to cite this passage
OpenAI, “Harness engineering: leveraging Codex in an agent-first world”, OpenAI (own site), 2026-02-11. https://openai.com/index/harness-engineering/ Archived at https://the-ai-record.com/passages/openai-escalate-human-only-when-judgment-required-2026
More from the same source
building and shipping an internal beta of a software product with 0 lines of manually-written code
OpenAIPublished by the organization itself (post), no named author
Post: Harness engineering: leveraging Codex in an agent-first world, OpenAI (own site), February 11, 2026.
- 100% ai code share, five months to 2026-02, one internal product. Level not stated
roughly 1,500 pull requests have been opened and merged with a small team of just three engineers driving Codex.
OpenAIPublished by the organization itself (post), no named author
Post: Harness engineering: leveraging Codex in an agent-first world, OpenAI (own site), February 11, 2026.
- 1,500 agent prs, five months to 2026-02. Level not stated
They pull review feedback, respond inline, push updates, and often squash and merge their own pull requests.
OpenAIPublished by the organization itself (post), no named author
Post: Harness engineering: leveraging Codex in an agent-first world, OpenAI (own site), February 11, 2026.
This translates to an average throughput of 3.5 PRs per engineer per day, and surprisingly the throughput has increased as the team has grown to now seven engineers.
OpenAIPublished by the organization itself (post), no named author
Post: Harness engineering: leveraging Codex in an agent-first world, OpenAI (own site), February 11, 2026.
- 3.5 prs merged per engineer, per engineer per day, five months to 2026-02
Something wrong? Report a correction. JSON: /api/v1/passages/openai-escalate-human-only-when-judgment-required-2026.json