OpenAI President Greg Brockman details 72-hour crisis and AI-driven development
In a wide-ranging interview, Greg Brockman outlines the near-collapse of OpenAI following Sam Altman’s dismissal, the creation of a backup entity, and the company’s evolving approach to artificial general intelligence in a compute-constrained environment.
Greg Brockman, co-founder and president of OpenAI, has provided a detailed account of the internal crisis that nearly dismantled the artificial intelligence leader, specifically focusing on the 72-hour period following the firing of chief executive Sam Altman. Speaking on the Farnam Street Blog’s Knowledge Project podcast, Brockman described the sequence of events that led to his own resignation on the same day as Altman’s dismissal and the subsequent creation of a backup entity known as “Phoenix”.
The backup company was designed at Altman’s house the morning after the initial board call, a move intended to ensure the continuity of OpenAI’s operations during a period of intense instability. Brockman noted that the crisis was ultimately resolved following a tweet from co-founder Ilya Sutskever, which helped to stabilise the situation. He also reflected on the company’s origins, noting that the original Napa offsite produced the three-step technical plan that has guided OpenAI for a decade.
Beyond the immediate crisis, Brockman addressed OpenAI’s structural evolution, explaining the strategic shift away from a pure nonprofit model. He detailed how the organisation had to abandon its initial structure to sustain its ambitions, a decision that has been central to its trajectory since he left Stripe in 2015 to help found the company. The discussion also touched on the broader context of the global AI race and the development of ChatGPT and GPT-5.
On the technical front, Brockman highlighted the significant integration of artificial intelligence into OpenAI’s own software development processes. When asked about the extent to which AI is used to write its own code, he remarked that it is hard to know what percent is not written by AI, indicating a substantial but unquantified reliance on automated coding tools. This shift underscores the company’s deep immersion in the very technology it produces.
The interview also covered recent operational changes, including the decision to stop showing reasoning traces in its model outputs. Looking forward, Brockman discussed the implications of a compute-constrained environment for access to artificial general intelligence. He suggested that such constraints could influence who gains access to AGI, raising questions about the future distribution of advanced AI capabilities and their impact on the workforce.
