Opening

Who is accountable when the agent gets it wrong, how many agents one human supervises, and how to move from “cost per token” to “cost per outcome.”