Non-engineers at OpenAI hit 40% Codex adoption while the app still showed them code

The sequence matters more here than the level: OpenAI shipped the Codex app for Mac in February and for Windows in March, and ChatGPT Work, which runs on the Codex harness, in July ^[10]. Between February and April the app still put code on the screen, and non-technical colleagues used it anyway, because it could research and produce a presentation, a document or a spreadsheet ^[12]. Adoption across non-engineering teams got close to 40% while the app was, in Orosz's description, hostile to those users ^[11]. Roughly 40 of the eventual 90 points, about 44% of the level reached, arrived before anyone made the interface comfortable for them ^[18].

Andrew Ambrosino, who leads Desktop at OpenAI, described the change at the level of artifacts. "The big theme of the past months has been that everything is now a coding agent. Whether the visible code is your output or not, agents write your artifacts," he said ^[14]. He put it this way: "Think of it like this: your entire life is via software. You have these powerful tools (agents) in your computer, and the ability to loop and reason and write code is the ability to do everything" ^[15]. Orosz dates the move from a nice-to-have tool to the backbone of pretty much everything at the company to the past year ^[3], and says almost all employees now use Codex and ChatGPT Work weekly ^[6].

Two caveats matter here. OpenAI's internal Codex is a lot more advanced than its external counterpart because it is plugged into pretty much every OpenAI system ^[13]. Then there is the bill. Engineers, researchers, finance colleagues and marketing staff there all work with an unlimited token budget ^[2].

Orosz's account says pull requests and code reviews at OpenAI need to be rethought ^[8], and the article does not describe the process that replaces them. A company that has to show an auditor who approved a change is being asked to adopt the agents first and rebuild the approval record afterwards.

The trade-off elsewhere is between metering and adoption. A token cap keeps spend forecastable and keeps the agent largely with people who already write code; the unmetered version is the condition under which finance, recruitment and legal moved their workflows across ^[5]. Setting that cap is this quarter's budget decision, and it determines how many non-engineers have the habit a year from now.