Post

is the product openai describe in this article… the codex macos app? they never name it. but the breadcrumbs are right there.

Steve Klabnik@steveklabnik.com

openai.com/index/harnes... i have been mildly obsessed with the ideas in here for the past month or so. incredibly validating, total victory: my suspicions are correct, actually. "This functions like garbage collection" is a sentence I have said out loud. craziness

the article says: first commit late august 2025, team of 3 engineers, “internal daily users and external alpha testers.” the codex desktop app launched february 2, 2026. five months later. simon willison confirmed he had preview access days before. i bet lots of others did too.

the article describes wiring chrome devtools into the agent runtime for dom snapshots, screenshots and video. the codex app is an electron app, so chromium.

altman apparently said (venturebeat): “this is the most loved internal product we’ve ever had.” the article: “used by hundreds of users internally, including daily internal power users.” hm..

the companion blog post from feb 4 (“unlocking the codex harness: how we built the app server”) describes the desktop app as a client of the app server protocol. published 3 days after the app launched. 7 days before the harness article.

the codex agent core is written in rust and predates this experiment. the rust rewrite dropped june 2025, before the august start date of “harness” so if it’s this, it would just be the electron frontend and all surrounding infrastructure - not the agent brain itself. lots of UI talk in the article

thanks for reading my first time investigative journalism, on an incredibly bland subject. if i’m at all right: openai used codex to build the codex app. and it’s out in the world now.

is millions of lines of code “too much” for the codex desktop app? i haven’t used it. nor have i worked on a comparable app

0 reposts0 likes

No replies yet.