How do you govern work that:
- moves faster than you can manage;
- makes stupid mistakes;
I learned this the hard way once already. My cybersecurity company, Kahuna, grew so quickly that Deloitte's Fast 50 named us the Sustainable Growth award winner. I believe it was 2006; interesting times. I sold Kahuna many years later.
Under the hood of that company, chaos exploded. We did not die. We survived; we hit serious growth pain, cut size, and watched good people leave because they were unhappy.
The fundamentals were not structured to carry that growth. Advisors taught me that later. If I had known them before the explosion, that company could have become a serious one. I learned the lesson. I am doing something else now.
That something else is building a company with autonomous, AI-powered agents run by my adult sons and me. Building a legacy with them, helping them learn what is not taught in their formal education.
With only fifteen agents, it already feels like a CEO with a thousand people and no middle management. The work packages finish in ten minutes, or thirty seconds, or two hours. You hold five packages at once. You become the bottleneck. You forget what you asked. The work exists somewhere, but the company record does not. Frontier model or cheap model, they do not behave like a hire. Model memory has gotten better over the last few months, but they can still be unbelievably stupid. They can follow rules until they do not.
The biggest remaining problems are managing fast growth, 7*24 availability, output speed, and scale. And brainstorming about governance structures with models trained on the corpus of human governance is both productive and unproductive.
Here is the difference from every board I have sat with or in. A classical board meets monthly and receives reports written by the company, which are mostly financial. Slow, and mostly produced by the same people they oversee.
An agent can read the size of a Bible in a minute and can keep it in its active memory. The P&L, or every ticket, every deploy, every customer conversation the company produced today. That means a board can be as informed as management. Sometimes more. That is the opening for a different governance structure we have built.
I used to be in the loop. I defined the speed. I do not want a bug report when another agent can fix the bug. I want a call when the plan is drifting, or when we have to go left or right. On the loop, not in it.
We at Zamforge have been building that model. It is early. It is how you stay responsible after you can no longer stay involved.
Rhett Oudkerk Pool · Originally posted on LinkedIn