Harness engineering: leveraging Codex in an agent-first world

An OpenAI engineering post on building a product with Codex and no hand-written code.

Author
Ryan Lopopolo, Member of the Technical Staff, OpenAI
Published
February 11, 2026
Kind
Company engineering blog post reporting one internal project
Original
openai.com/index/harness-engineering/
Subject
Harnesses and agent loops
Themes
Handing over control

What it is

Ryan Lopopolo writes for OpenAI about a five-month experiment. The post sets out what the team learned about designing environments for agents.

What it says

What evidence it gives

The evidence is the team's first-hand account of one project. Figures such as pull request counts and a one-tenth time estimate come from the team. The post includes a docs layout and a layer diagram. It offers no outside measurement.

Interests and limits

OpenAI wrote the post and makes Codex. The time saving is an estimate, and there was no comparison build. The product was internal, with alpha testers. The post says it does not know how coherence holds up over years. The direct page load was blocked by a JavaScript check. This record uses a public Internet Archive copy of the page.

Connections

These links are this wiki's own reading, not claims the sources make.