The former Openai engineer explains what it really is like to work there
Three weeks ago, an engineer named Calvin French-Owen resigned from the company, working on one of Openai’s most promising new products.
He just made public Charming blog posts What was it like working there for a year, including the sleepless sprint to build codexes? that’s right Openai’s new coding agent It’s a competition Cursor-like tools And the human race Claude Code.
France’s Owen said he didn’t leave because of “drama,” but wanted to return to the startup founder. He was a co-founder of the Customer Data Startup Segment. Buyed by Twilio in 2020 for $3.2 billion.
Some of his reveals about Openai culture will surprise anyone, but other observations counter some misconceptions about the company. (He could not immediately contact him for comment.)
Rapid Growth: Openai grew from 1,000 to 3,000 in the year he was there, he wrote.
LLM model makers certainly have such a reason for employment. It is the fastest growing consumer product to date, and its competitors are also growing rapidly. In March, it said ChatGpt had over 500 million active people Climb immediately with the user.
chaos: “How we communicate as a company, reporting structure, how we ship products, how we manage people, how we manage people, how we hire processes, and more: when we scale quickly, everything breaks,” writes France Owen.
Like a small startup, the people there are still empowered to act on their own ideas. But that also means that multiple teams are replicating their efforts. “You’ve probably seen half a dozen libraries for queue management, agent loops, etc,” he provided as an example.
Coding skills also range from veteran Google engineers who write code that can handle 1 billion users to unfinished doctoral degrees. This, coupled with the flexible Python language, means that the central code repository, aka “backend monolith,” is a “small garbage dump.”
Things can break frequently and take excessive time to run. But top engineering managers are aware of this and are working on improvements, he writes.
“Starting Spirit”: Openai still doesn’t seem to know that it’s a huge company and is supposed to run entirely on Slack. He observed that in the early days of Facebook it was very similar to the fast and split meta. The company is also full of recruitment from Meta.
France’s Owen explained how a senior team of about eight engineers, four researchers, two designers, two commercial staff and one product manager created and launched the codex in just seven weeks.
But launching it was magic. Just turn it on and they got users. “I’ve never seen the product grow anytime soon, just appearing in the left sidebar, but that’s the power of ChatGpt.”
Secret Fish Bowl: ChatGpt is a highly scrutinized company. This led to a culture of secrets to crack down on leaks to the public. At the same time, the company monitors X. The post is spreading, and Openai sees it and probably responds to it. “A friend of mine joked, ‘This company runs in a Twitter vibe,” he wrote.
The biggest misconception: France’s Owen implies that the biggest misconception about Openai is that it is not as concerned about safety as it should be. Certainly, many AI safes, including the former Openai employees criticize the process.
There are destinies who are worried about the theoretical risks to humanity, but internally they focus more on practical safety, such as hate speech, abuse, manipulation of political bias, manipulation of biowarepon, self-harm, and rapid infusion. Openai doesn’t ignore the potential long-term impact, he wrote. While some researchers are looking at them, they recognize that hundreds of millions of people use LLM today for everything from medical advice to treatment.
The government is watching. Competitors are watching (and Openai is looking at their competitors in return). “I feel the stakes are really high.”