Reflections
In this January 2025 essay, Sam Altman looks back on OpenAI's first nine years, from ChatGPT's surprise launch to his abrupt firing by the board. He argues that building a company in a brand-new category has no handbook, and that the work forced lessons on governance, scale and changing tactics.
We link to the original and never republish it. All credit to Sam Altman.
Key takeaways
- Altman says the ChatGPT moment was a surprise: the team always expected a tipping point but did not know it would be a demo built around how developers already enjoyed talking to the model.
- When a category is new, Altman says no one can tell you how to do it, so expect two steps forward and one step back, with mistakes corrected as you go.
- Altman calls his firing a failure of governance by well-meaning people, himself included, and says he learned a board needs diverse viewpoints, broad experience, trust and credibility.
- Altman says turnover and colleagues leaving to compete are close to unavoidable when numbers go up by orders of magnitude every few months.
- Altman says OpenAI's vision stays fixed while tactics evolve: they did not expect to need to build a product company or raise such a large amount of capital.
Who it's for
Founders running a fast-scaling company in a new category, especially those thinking about board composition and handling a crisis.
Why it's worth your time
It is a candid first-person account of the stress, mistakes and gratitude behind a very fast-growing company, and the lessons on governance and changing tactics hold regardless of what happens with AI.
More from Sam
All from Sam Altman →More on founder stories
All 275 →The best founder writing, in your inbox
Hand-picked essays, talks and decks from the founders who built the companies. Free, and you can unsubscribe any time.
86 minutes of paid-UGC alphaHector Serrano · Wispr Flow
ChatGPT Should Have Killed His AI SaaS. Instead, He's Nearing $10MAlon Talmor · Mosaic AI
Databricks' Ali Ghodsi Never Wanted to Be CEO. Now He's Among the BestAli Ghodsi · Databricks