The Rich-vs-King Problem Has a New Dimension
Wasserman's The Founder's Dilemmas tracks thousands of startups and finds that founders who optimize for control tend to build smaller companies, while those who cede it tend to build more valuable ones. That tradeoff always existed. What changes in an AI-native company is that control is no longer just about equity or board seats. Control over the eval loop, over where human judgment sits, over which decisions agents make autonomously — these are structural choices baked in at founding. A cofounder who cannot agree on those boundaries early will not agree on them later, when the stakes are higher and the system is already running.
Equal Equity Is the Wrong Default
Wasserman's data is unambiguous: equal splits between cofounders, especially when agreed on quickly to avoid discomfort, predict conflict. The logic is that equal equity papers over unequal contribution and unequal roles. In an AI-native company, roles are genuinely asymmetric from day one. Someone owns the agent architecture and the eval harness. Someone owns customer judgment and distribution relationships. Someone owns the accountability layer where humans review consequential outputs. Pretending those contributions are equivalent because the conversation is awkward is exactly the trap Wasserman describes — and in this context it leaves the operating loop without a clear owner.
The Least Reversible Choices Are Structural
What makes Wasserman's research durable is the emphasis on irreversibility. Early cofounder agreements, vesting schedules, and equity splits calcify. AI-native companies have an analog: the design of the human-in-the-loop layer. Who holds judgment, how escalation works, where the company stops and defers to a human — these are not features to be added later. They are the operating system. Founding teams that defer this design because it feels like a product question rather than a people question will discover, as Wasserman's founders did, that the earliest unresolved tension becomes the most expensive one.
Founding Team as the First Eval Committee
Wasserman frames the cofounder relationship as a governance problem. AI-native founders can go further and treat the founding team as the first eval committee — the group that decides what good output looks like, where autonomy ends, and who owns the call when the agent is wrong. That function cannot be delegated to the model and it cannot be split equally without an agreed tiebreaker. The founding team either designs trust into the system deliberately or discovers its absence in production.
