What Is It Like to Work at OpenAI?

Tibo Sottiaux says working at OpenAI is bottoms-up energy from ex-founders, with trust to own choices and a quality bar high enough to hold launches back.

Decisions API came together quickly after the team saw a strong model and used constraint sampling to ship a different shape of the Responses API. It is faster and quite delightful for some cases. A Slack channel started with four people hacking on a weekend. People then got excited, added visual inputs, and pitched in with no fixed structure until a product shipped. The bar for quality still stops work that needs more baking. Launches that were maybe going to ship the same day were held and spaced out, with more coming the next week and the week after. People take on roles that channel the energy into productive work.

The autonomy comes with ownership. People are trusted to make great choices, then to fix a mistake quickly or learn from it. He can press reset whenever it feels right, stay close to the community, and skip an echelon of approvals. The miss he names is pushing teams to keep striving toward complexity when he could have inspired them to build things that are less complex.