Sam Altman

Sam Altman was a founding co-chair of OpenAI alongside Elon Musk and returned as its CEO in November 2023. His work spans startup leadership and the development of an AI product company: he describes ChatGPT as a collective response to developers already conversing with models through OpenAI’s API playground. His public views emphasize gradual deployment, human direction of agents, and AI tools that help people generate better ideas.[9][13][11][12][1]

At an OpenAI town hall reported by Latent.Space on January 29, 2026, Altman argued that AI should improve the quality of people’s thinking. Responding to Swyx’s question about better ideas, he said, “we think at the limits of our tools”. He proposed systems that examine a person’s past work and code to suggest useful or interesting next steps. Falling creation costs, he argued, could tighten feedback loops by letting people test ideas faster and discard weak ones sooner.[1]

Altman used Paul Graham as his example of a strong brainstorming partner: someone whose questions and suggestions leave him with new ideas. A proposed “Paul Graham bot” could be valuable even if a user rejected 95% of its suggestions. He also reported that scientists were finding nontrivial scientific progress from an internal version of GPT-5.2, and suggested that a different harness and training could help models generate product insights. These were proposals and his account of internal feedback, rather than an announcement of a finished brainstorming product.[1]

Altman also encouraged broader uses of coding agents. Latent.Space’s May 1, 2026 recap quoted him urging users to “try it for non-coding computer work” with Codex. Its May 2 recap quoted his expressed aim to “build tools to augment and elevate people, not entities to replace them”.[4][5]

Altman’s relationship with Graham predates these AI proposals. He participated as a founder in Y Combinator’s first funded batch in 2005 and later took over its leadership, providing the earlier career context for his town hall example.[10][1]

On February 21, 2014, Graham announced that Altman would become Y Combinator’s president starting with the next batch. Graham would continue office hours while Altman led YC. Graham said he had begun recruiting Altman after he became available in 2012 and that the effort took more than a year. His stated reason for the transition was YC’s need to grow; he and Jessica considered Altman best suited to that task.[10]

OpenAI’s December 11, 2015 introduction named Altman and Musk as its co-chairs.[9]

In Reflections, Altman recalls that developers enjoyed talking to models in the API playground. OpenAI decided to build a demo around that experience to show something about the future and improve its models’ quality and safety. Temporarily called “Chat With GPT-3.5,” it launched as ChatGPT on November 30, 2022. Altman describes its reception as a surprise tipping point that required the research lab to build a product company around the technology, with capital needs it had not originally anticipated.[11]

Altman recounts being unexpectedly fired on a video call while in a Las Vegas hotel room. He calls the episode “a big failure of governance by well-meaning people, myself included”, and says it taught him the importance of board members with diverse viewpoints and experience, as well as trust and credibility. He credits Ron Conway and Brian Chesky with extensive crisis assistance, including strategic advice and stopping him from making mistakes. OpenAI announced his return as CEO on November 29, 2023, alongside a new initial board.[11][13]

Altman advocates iterative deployment: gradually releasing AI into the world, learning from its use, and giving society time to adapt. In Reflections, he argues that safety and alignment research should draw on feedback from real applications. This is his rationale for deployment, rather than a claim that release itself guarantees safety.[11]

In Three Observations, Altman imagines software agents as relatively junior virtual coworkers that could eventually handle tasks lasting up to a couple of days. He attaches substantial limits to that forecast: they would need human supervision and direction, perform unevenly, and lack the biggest new ideas. His scenario then asks what large numbers of such agents could accomplish across knowledge work.[12]

Altman’s economic argument connects model intelligence with training compute, data, and inference compute. He argues that falling costs can expand use and that increasing intelligence can justify further investment. He also warns that broadly shared benefits and equality do not follow automatically from technological progress: the balance between capital and labor could require intervention. He considers possible individual compute budgets and cheaper access to intelligence, while acknowledging tradeoffs between safety and individual empowerment.[12]