Greg Brockman walks through the 72 hours that almost killed OpenAI
In a rare interview on The Knowledge Project, the OpenAI co-founder recounts quitting within hours, sketching a backup company, and why they stopped showing reasoning traces.
By Ryan Merket · Published · Updated
Why it matters
Founder accounts like Brockman's are rare and useful. They surface how leaders actually make calls on structure, safety, product defaults, and contingency under stress, which are the levers that will shape who builds and accesses the next wave of AI.
Greg Brockman, OpenAI co-founder and President, uses a new episode of The Knowledge Project to tell the inside story of the three days after Sam Altman (@sam) was fired and the contingency plan he and Altman drafted the next morning.
The founder in the room
Brockman is not a generic executive parachuted into AI. He was the first engineer at Stripe and later its CTO before leaving in 2015 to help start OpenAI (career timeline). In the conversation with host Shane Parrish, he rewinds to OpenAI's origin: an early Napa offsite where the team sketched a three-step technical plan they have followed for a decade, and he explains why OpenAI ultimately moved away from a pure nonprofit structure (all per the episode page and transcript).
The 72-hour crisis
The heart of the episode is Brockman's chronology of the weekend Altman was removed. He recounts where he was when the board call came, why he quit the same day, how a backup company called "Phoenix" was outlined at Altman's house the next morning, and the moment Ilya Sutskever's tweet changed the trajectory. These are Brockman's own recollections, shared in the podcast and linked clips like "Sam Altman's Firing" and "Breakthrough Moments at OpenAI."
Product and policy calls
Brockman also looks forward. He addresses whether we are in a global AI race, and how much of OpenAI's code is now authored by AI. The show page summarizes his view as: "it's hard to know what percent is not." He explains why OpenAI stopped showing reasoning traces in ChatGPT, and how a compute-constrained world could shape who gets access to advanced systems.
That context sits alongside an intense shipping cadence. OpenAI's site highlights recent updates like GPT-5.5 and ChatGPT Images 2.0, and research posts spanning content provenance and a model disproving a discrete geometry conjecture.
Why governance mattered
Brockman does not relitigate every board detail, but he does make one point plain: structure and mission are not abstractions when the stakes are this high. He describes the rationale for abandoning a pure nonprofit model and the governance mechanics around OpenAI's current setup, framed by the organization's founding goal to build safe, beneficial AGI. The episode page characterizes OpenAI today as the company behind ChatGPT and GPT-5, with the nonprofit Foundation governing the public benefit corporation that operates the products.
The operator's takeaway
For founders and operators, this is not just OpenAI lore. It is a case study in contingency planning under pressure, in choosing product defaults like reasoning visibility, and in designing organizations that can survive a crisis. Brockman has been writing code and running teams since Stripe's earliest days; here he is candid about how those instincts carried through OpenAI's wildest weekend and into decisions that will shape the next phase of AI.
Listen on YouTube, Spotify, Apple Podcasts, or read the transcript.