Calvin French-Owen, former OpenAI technical staff, shares his post-departure reflections

Jul 16, 2025

On the Culture

The culture is very bottoms-up and meritocratic, where the best ideas tend to win regardless of politics.
There's a strong bias to action, where small teams or individuals can start projects without asking for permission.
The company changes direction quickly and goes all-in once a decision is made.
Due to intense public scrutiny, OpenAI is a very secretive place.
The stakes feel high, creating a more serious atmosphere than one might expect.
The company scaled incredibly fast, from ~1,000 to over 3,000 people in one year.
All internal communication runs on Slack; there is no email.

The Engineering Environment

The company uses a giant monorepo that is mostly Python.
Everything runs on Azure.
There is a significant engineering talent pipeline from Meta, and the infrastructure feels reminiscent of early Meta.
Code wins over central planning committees, which leads to a bias for action but also duplicate codebases.
The idea of chat messages and conversations is a deeply baked-in primitive in the codebase.

The Codex Launch Sprint

The entire product was built from the first line of code to a full public launch in just 7 weeks.
Described as the hardest they had worked in nearly a decade, with late nights and weekend work.
The core team consisted of ~8 engineers, ~4 researchers, 2 designers, 2 GTM, and a PM.
In the 53 days after launch, Codex had publicly generated 630,000 PRs.

Other Insights

GPU costs are so massive that nearly every other expense is a rounding error by comparison.
The company is "frighteningly ambitious," aiming to compete in dozens of arenas beyond its consumer app.
OpenAI pays close attention to Twitter; viral tweets can influence internal consideration.

marvin's notes