A former OpenAI engineer describes what it’s really like to work there

Three weeks ago an engineer named Calvin French-Owen, who worked on one of the most promising new products from OpenAi, resigned from the company.

He just published one Fascinating blog post On what it was like to work there for a year, including the sleepless sprint to build Codex. That is the new coding agent from OpenAi who competes with tools such as Cursor and Anthropic’s Claude Code.

Frans-Owen said he was not leaving because of any ‘drama’, but because he wants to become a startup founder again. He was co-founder of the startup segment of the customer data, which was purchased by Twilio in 2020 for $ 3.2 billion.

Part of what he revealed about the OpenAi culture would not surprise anyone, but other observations fight some misconceptions about the company. (He could not be reached immediately for comment.)

Rapid growth: OpenAi grew from 1,000 to 3,000 people in the year he was there, he wrote.

The LLM model maker certainly has reasons for such recruitment. It is the fastest growing consumer product ever and its competitors are also growing fast. In March it said that Chatgpt had more than 500 million active users And climb quickly.

Chaos: “Everything breaks when you quickly scale that: how to communicate as a company, the reporting structures, how to send product, manage and organize people, the recruitment processes, etc.,” wrote French-Owen.

Just like a small startup, people there are still authorized to act according to their ideas with little to no bureaucracy. But that also means that several teams are duplicating efforts. “I must have seen half a dozen libraries for things such as queue management or agent loops,” he offered as examples.

Coderting skills also varies from experienced Google -Seniurs who write code that a billion users can handle, to a new PhD students who do not. This, in combination with the flexible Python language, means that the Central Code Repository, also known as ‘the back-end Monolith’, is ‘a bit of a landfill’, he described.

Stul often breaks or can take excessive time to run. But top technical managers are aware of this and work on improvements, he wrote.

“Last spirit”: OpenAi does not seem to know that it is a huge company, up to on the way to Slack. It feels very much when Meta from the Move-Fast-en-Things in his early Facebook years, he noticed. The company is also full of recruitment from Meta.

French Owen described how his senior team of about eight engineers, four researchers, two designers, two go-to-market employees and a product manager built and launched Codex in just seven weeks, from start to end, with almost no sleep.

But launching was magic. Simply by turning it on, they have users. “I have never seen a product get so much immediate rise, just by appearing in a left sidebar, but that is the power of Chatgpt.”

Secretive Fishbowl: Chatgpt is a highly examined company. This led to a culture of confidentiality in an attempt to clamp on leaks to the public. At the same time, the company looks at X. If a post goes viral there, OpenAI will see it and possibly respond to it. “A friend of mine joked:” This company runs on Twitter -Vibes, “he wrote.

Greatest misconception: Frans-Owen implied that the biggest misconception about OpenAi is that it is not as worried about safety as it should be. Certainly many AI safety people, including former OpenAi employees, criticize the processes.

Although Doomsayers are concerned about theoretical risks for humanity, there is more internal focus on practical security such as “hateful speech, abuse, manipulating political prejudices, making bio-weapons, self-harm, fast injection,” he wrote.

OpenAi does not ignore the potential effects in the long term, he wrote. There are researchers who look at them and it is aware that hundreds of millions of people today use his LLMS for everything, from medical advice to therapy.

Looking governments. Watching competitors (and OpenAi is watching competitors in exchange). “The bet feels really high.”