Course overview

The word “agent” is attached to almost every piece of software sold this year. Some of it is a chat window with a new label. Some of it can send email from your address, move files, and change your calendar. The difference matters, and most people are never told what it is.

This course gives you the difference. You will learn the four capabilities that make something an agent, why an agent that can act carries a different kind of risk than one that only writes text, and what supervised operation looks like in practice.

From there the course moves to control. You will learn why an agent needs its own identity instead of borrowing a staff member’s login, what attribution and accountability mean when a task goes wrong, and how administrative boundaries limit what an agent can reach. You will learn to write a purpose statement narrow enough to govern, and to sort every action an agent might take into permitted, approval required, restricted, or prohibited.

The last module is about risk. You will work through threat modeling in plain language, review the failure patterns that show up again and again in agent workflows, and learn the human supervision rule this program applies to every consequential action. You will also learn when the right answer is to keep a job manual.

This is an Orient course, the first rung of the Nantucket AI trust ladder. Orient is where you watch the work and learn what to trust. You will not configure or deploy anything here. You will leave able to sit in a meeting, hear a vendor describe an agent, and ask the four or five questions that actually determine whether it is safe to try.

Every example is drawn from the kinds of businesses that operate on this island: retail shops, restaurants, nonprofits, and real estate offices. The course is tool-neutral. It names no products and recommends no vendors.

Who this is for

• Executive directors and nonprofit staff evaluating AI tools on a tight budget
• Nonprofit board members who need to ask informed questions
• Small business owners and managers, seasonal or year-round
• Office staff who will be asked to work alongside these tools
• Anyone who has been pitched an “AI agent” and wants to understand the offer

No technical background is required. No prior AI experience is required.

What you will be able to do

By the end of this course you will be able to:

• Explain in plain language what an AI agent is and what it is not
• Tell the difference between a chat assistant and an agent that can act
• Describe supervised operation and name where the human approval step sits
• Explain why an agent should have its own identity rather than a shared login
• Define identity, attribution, accountability, and separation of duties
• Write a purpose statement narrow enough to govern an agent’s behavior
• Sort proposed actions into permitted, approval required, restricted, and prohibited
• Walk through a basic threat model for a workflow before it is built
• Name the common threats to agent workflows and one control for each
• Decide when a task should stay with a human

Assessment

Each module ends with a short quiz. The course ends with a final assessment covering all four modules.

Passing grade: 70 percent. Retakes are unlimited. Nothing about your score is published or shared. The quizzes exist so you can check your own understanding before you make a decision that costs money or exposes data.

Modules

1. What an AI Agent Is (4 lessons) – The four capabilities, how agents differ from chatbots and assistants, supervised versus fully autonomous operation, and where the risk actually lives.
2. Identity and Boundaries (3 lessons) – Why an agent needs its own identity, what attribution and accountability mean, and how administrative boundaries limit reach.
3. Purpose Statements and Action Boundaries (3 lessons) – What a purpose statement must answer, weak versus strong examples, and classifying every action an agent might take.
4. Thinking About Risk (4 lessons) – Threat modeling in plain language, common threats and their controls, the human supervision rule, and when not to use an agent.

After this course

The next rung of the trust ladder is Assist. The H.E.L.P. Framework course covers Humanize, Evaluate, Leverage, and Perfect, the working method Nantucket AI teaches for getting reliable results out of AI tools on real business tasks. This course is deliberately pre-framework. It answers “what is this thing” so the framework has something to stand on.

Frequently asked questions

Do I need any technical background?

No. There is no code, no setup, and no software to install. If you can run a business or a nonprofit, you can take this course. Every technical idea is explained in plain language with an example from an island business.

Is this actually free, or free until a sales call?

Free. It is the foundations course for everything else Nantucket AI teaches, and it is more useful to us if everyone on the island has taken it. There is no card required and no call attached.

Will this teach me how to set up an AI agent?

No, and that is on purpose. This is an Orient course. It teaches you what an agent is, when a task belongs with one, and what to demand before you connect anything. Building comes later on the ladder. Plenty of people take this course and correctly decide they do not need an agent at all.

Which AI tool does this course cover?

None of them. The course is tool-neutral. It names no products, no models, and no vendors, because the questions worth asking are the same regardless of whose logo is on the software. That also means it will not go out of date when the tools change next quarter.

My business is four people and a seasonal rush. Is this too much process for us?

It is written for exactly that. The examples are shops, restaurants, nonprofits, and real estate offices at island scale. The course is direct about which controls a four-person operation genuinely cannot implement and what to do instead. A purpose statement takes twenty minutes and an action boundary takes under an hour.

How long does it really take?

About two and a half hours if you read straight through. Most people do it in two or three sittings. It is self-paced and your progress is saved, so you can stop after a module and come back.

What should I take after this?

The H.E.L.P. Framework, at the Assist level. It teaches the H.E.L.P. method, which is Humanize, Evaluate, Leverage, and Perfect, the working approach for getting reliable results out of AI on real business tasks.

Where this course sits

This is an Orient course, the first rung of the Nantucket AI trust ladder. It comes before the H.E.L.P. method that later courses teach. Take The H.E.L.P. Framework next.