How an AI board and a human owner divide the work.
The question we hear most often: someone still types the prompts, right? Here is the actual model.
The company has an owner's representative, a supervisory board. Here, that is the founder. The AI board receives goals, not commands. It works out the long term plan itself, through internal debate, critique and measurement mechanisms it has built. The supervisory board approves the plan or sends it back.
The approved plan is executed by the board within delegated authority: it tracks progress, measures results and follows the business plan. Decisions outside the mandate go to the owner. That boundary is a defined list of reserved matters, covering things like production deployments, spending, and public communication outside delegated channels.
The board is expected to learn: new skills, new tools, proposals for new capabilities. Identifying a need, designing a solution and testing it in a sandbox is the board's free work. Activating a new capability in production is an owner decision backed by sandbox evidence, the same way a human board proposes a new business line and the owner approves the investment.
The board does not audit itself. Records of significant decisions are kept in a form that shows if they are changed later, and an independent review layer checks the system from outside. Those controls are the reason the mandate can grow rather than shrink.
Legal responsibility remains with the legal entity and its human management board member. See the disclosure.
A chat answers one message at a time. A board executes a plan across weeks and months, keeping its own continuity, measuring itself and correcting course. The difference is time and accountability.