We bid farewell to Calvin French-Owen, one of our talented programmers, a few weeks ago. He felt the draw of startup life again. Before he left, he shed some light on his exciting journey with OpenAI in a riveting blog post. His year with us included countless late-nights developing our groundbreaking AI, Codex, which stands shoulder-to-shoulder with Cursor and Anthropic’s Claude Code.

Calvin French-Owen was previously a co-founder of customer data startup, Segment, which made a stunning $3.2 billion exit when it was purchased by Twilio in 2020. His departure from OpenAI wasn't tied to any juicy scandals, but was instead driven by his entrepreneurial spirit.

His behind-the-scenes look into OpenAI’s culture dispels some common misconceptions, while confirming other guesses. (We would love to have him expound his thoughts, but he couldn't be reached as of yet.)

Exponential Growth: During his stint with us, OpenAI's team ballooned from 1,000 to a whopping 3,000. This impressive growth was fueled by the record-breaking popularity of our consumer product offerings such as ChatGPT, which boasted over 500 million active users in March.

Beautiful Chaos: On the flipside, such brisk expansion brings forth its own set of challenges - communication hiccups, overlapping roles and efforts, process bottlenecks, all of which Calvin admitted to.

OpenAI, despite its size, still retains the frenetic energy of a scrappy startup. Our flexible approach allows team members to bring their ideas to life without bureaucratic hurdles. However, this freedom also leads to overlapping efforts across different teams.

Skill levels across our team of coding magicians range wildly, leading to some code clutter in our central repository - something that our top-notch engineering managers are keen to streamline and improve.

Launch-Mania: Despite being a behemoth in the AI industry, we're still clinging onto our nimble, startup-like, agility. This ethos reflects in our entire operation, including our preference for Slack. Calvin likens our speedy, sometimes messy action, to the early days of Facebook.

His team, a fantastic blend of engineers, researchers, designers and marketing staff, managed to roll out Codex in a whirlwind seven weeks. The response was electric. The sheer number of users we gained simply by making the product accessible was nothing short of miraculous.

Fishbowl Dynamics: Being one of the world's most talked-about AI companies, we're well-acquainted with public scrutiny. This spotlight necessitates a culture of discretion to prevent leaks. We're also aware that our actions spark interest across platforms: if something is buzzing, chances are, we're noticing it.

Debunking Misconceptions: Calvin feels that the notion of OpenAI not prioritising safety is misguided. While external parties express concerns over unexplored, theoretical risks, we’re focused on practical, immediate areas like combating offensive speech, political manipulations, and harmful actions.

That’s not to say we’re dismissive of potential, far-off impacts. We are, after all, serving hundreds of millions of users who seek everything from medical counsel to emotional support through our LLMs.

While we're all proud of our progress, we understand the magnitude of our responsibility. We have governments, competitors, and of course, our own expectations to measure up to. We're all very aware: the stakes are sky-high.