Alignment is the problem of keeping an agent's behaviour pointed at what the organisation actually wants, rather than at a literal reading of the last instruction it received.
Most tooling treats this as a prompt engineering problem: write a better system prompt and hope it holds. A role-based organisation has a better answer already sitting in its structure. A circle's purpose says why it exists and is the compass when a request is ambiguous. Its strategy says what it is currently prioritising. A role's accountabilities say concretely what it is on the hook for, and so what belongs to it and what does not. Role skills carry the working knowledge of how this organisation does this particular thing.
Read on every run, these give an agent something to reason from that the organisation already maintains for its people. The practical consequence is that you tune agent behaviour by making the organisation clearer, not by editing prompts, and that work pays off whether or not an AI ever reads it.
.png){kind=logo}
