If an AI agent is going to act inside your systems, it has to act as somebody. The most common shortcut is to let it use a staff member's login. This module explains why that shortcut causes problems you will not notice until you are trying to investigate something. You will learn what a dedicated agent identity gives you, how identity, attribution, and accountability differ, and how administrative boundaries limit what an agent can reach. These are the questions to ask any vendor before you connect their product to anything.

An agent with a vague job is impossible to govern. If nobody wrote down what it is for, then every question about whether it should have done something becomes an opinion. This module teaches you the two documents that fix that. A purpose statement says what the agent is for, in writing, narrowly enough that a stranger could tell whether a given task is inside or outside its job. An action boundary sorts every action the agent might take into four categories. You will learn to recognize a weak purpose statement in about ten seconds, and to write a strong one.

Risk work has a reputation for being either paranoid or performative. Done properly it is neither. It is a short, structured conversation you have before you build something, and it usually takes under an hour. This module teaches you that conversation. You will learn a seven-step method for thinking through how a workflow could fail, the failure patterns that come up repeatedly in agent work, the human supervision rule this program applies to every consequential action, and how to recognize the workflows that should stay with people.