Note ·
Small, focused, high-trust
In a 2020 vision memo, written after Microsoft’s investment, Sam Altman warned his own company against growing too large. Karen Hao quotes it in Empire of AI (chapter 11): “what has worked for us so far is being small, focused, high-trust, low-bullshit, and intense.” Too many people and too much bureaucracy, he wrote, “can easily kill great ideas or result in sclerosis”, and OpenAI could fulfil its mission “with a surprisingly small number of great people.”
He meant it. Hao reports that in late 2022, after ChatGPT, when every team was stretched thin and managers begged for head count, Altman kept repeating the line and wanted to add no more than about a hundred people. Other executives wanted closer to five hundred. OpenAI had about 300 people in October 2022, close to 600 by the middle of 2023, and nearly 800 when the board fired Altman that November.
I have made the same observation in every large organisation I have worked in or with, public or private. A group of people has to spend work on holding itself together, and that share grows faster than the group. Fred Brooks put the arithmetic in The Mythical Man-Month: every pair of people is a possible line of communication, and the number of pairs grows with the square of the headcount.
| People | Pairs who may need to talk |
|---|---|
| 5 | 10 |
| 10 | 45 |
| 50 | 1,225 |
| 500 | 124,750 |
| 5,000 | 12,497,500 |
Nobody talks to everyone, so the organisation adds structure: managers, meetings, processes, reporting lines, approval steps. Each is reasonable. Together they become the leviathan, a body whose main occupation is keeping itself intact, whether it is a ministry or a multinational. Its slowness is the price of its coherence.
What interests me is that the arithmetic may be changing. If a small team with agents can do what used to take a department, the number of people a mission needs drops, and with it the overhead. I wrote about that from the side of the individual in power in the good sense and citizen entrepreneurs. Altman’s “surprisingly small number of great people” reads differently in 2026 than it did in 2020.
Small has its own failure mode, and Hao’s account shows it too: teams stretched so thin that the people in them burn out. Small, focused, and high-trust is a discipline, and OpenAI could not keep it. Neither, as far as I know, has anyone else at that scale.
Related: A billion without a plan, and my review of the book, An Empire of People.
References
- Karen Hao, Empire of AI (2025), chapter 11, “Apex”.
- Frederick P. Brooks Jr., The Mythical Man-Month (1975), chapter 2.