How We Used Codex to Ship Sora for Android in 28 Days

Twenty-eight days. That's how long it took OpenAI to take Sora, its video-generation app, from a standing start to a live release on Android. No massive engineering org, no months of sprint planning. Just a small team leaning hard on Codex, OpenAI's own coding agent, to do the grunt work that usually eats up calendars.

The approach wasn't about letting the AI write the whole app and walking away. According to OpenAI's own account, the team used Codex for planning out features before a line of code existed, for translating strings and interface text across languages, and for running coding tasks in parallel rather than one engineer chewing through a to-do list top to bottom. That parallelism seems to be the real trick here — instead of a linear pipeline where each step waits on the last, multiple pieces of the app moved forward at once, with Codex handling the repetitive, mechanical parts while humans made the judgment calls.

This matters beyond Sora itself. Mobile app development, especially porting an iOS-first product to Android, is notorious for dragging on for months because of fragmented devices, translation work, and testing across configurations. Compressing that into four weeks with a lean crew is the kind of case study that other engineering leads will point to when they're arguing for AI coding tools in budget meetings. OpenAI is, unsurprisingly, using its own product to prove its own product works.

There's also a quieter signal in here about what OpenAI thinks Codex is for. This isn't a demo about generating a snake game or fixing a bug in five seconds. It's about running an actual shipping process — planning, localization, execution — with an AI woven into each stage rather than bolted on at the end. If that pattern holds for other teams, the 28-day number won't be the interesting part for long. What'll be interesting is how many other companies quietly start moving at that pace and never say a word about how they did it.