#define CTO OpenAI — Greg Brockman

Sequel to #define CTO, written two years later after cofounding OpenAI. Where the Stripe post was about discovering a role inside an existing org, this is about finding a problem bigger than your role, assembling the right people around it, and discovering what you can uniquely contribute.

Formation: social before technical

OpenAI started as a dinner discussion. Brockman's job before launch: build the founding team. Key move — chaining through networks, then using endorsements to convert skeptics. John Schulman's endorsement made Andrej Karpathy and Wojciech Zaremba pay attention. Mission resonated immediately; the challenge was making the unformed organization feel real.

Engineering as research bottleneck

The core thesis: ML systems = ML core (algorithm, few chapters of a textbook) + large surrounding software engineering. Engineering enables research. Each incremental engineering improvement makes problems easier for models and can push research over the finish line.

Two examples:

Gym: John Schulman estimated year-end delivery. Brockman shipped in two months by clearing his schedule and going heads-down. Pattern from Stripe: focus single-mindedly, working from wake to sleep, which raises the quality bar and inspires others.
Universe (VNC-based agent environment): blocked by Pong taking 50 hours to train → set derisking goal of 1 hour (achieved; now 10 minutes). Engineering bottleneck became the research question: "no one had ever tried benchmarking a single agent across thousands of games."

"At OpenAI, even the slogwork is fundamental."

Soft/hard launch date pattern

Always set both a soft and a hard launch date, spaced 1-2 weeks apart. Brockman has never hit a soft launch date but has never missed a hard one.

Block time for deep work

When in coding mode: a single meeting kills morning or afternoon productivity; meetings in both periods make evening coding impossible. Rules: restrict meetings to early morning or right after lunch; no more than 3 meetings/day; no back-to-back meeting blocks more than every other day.

The role found

At OpenAI, the CTO role naturally aligned with what Brockman loves most — coding — because engineering is the actual bottleneck. Not a role designed from the outside; discovered by finding the problem first, building the team, and letting the contribution emerge.

"People remain my focus: the journey to this point was about a social story, not a technical one. Going forward, this role is only sustainable because I'm part of a fantastic team willing to jointly handle the 'trucks or telephones.'"