What an Agent Should Not Do
Imagine you have a new puppy. The puppy is very eager to help! It runs to the door when someone knocks. It brings you things it finds on the floor. It barks when it hears a strange noise. But if you do not teach the puppy some rules, it might chew up your favorite book, bark in the middle of the night, or bring you something it should not. Rules are not mean — they help the puppy do the right thing. AI agents are the same way. Setting clear rules for what an agent should NOT do is one of the most important jobs of a good boss.
Agents Need Limits, Not Just Instructions
When you give an agent a task, you tell it what TO do. But a complete set of instructions also needs to say what NOT to do. For example: You might tell an agent: help me write emails. But without a limit, it might also delete old emails, unsubscribe from things, or reply to messages you did not mean to answer. A better set of instructions includes limits: help me write draft emails — but do NOT send anything, do NOT delete anything, and do NOT reply to anyone without showing me first. Limits protect you. They keep the agent inside the boundaries of what you actually wanted.
Good instructions for an agent include what TO do AND what NOT to do. Setting clear limits keeps the agent safe, helpful, and under your control.
Here are some limits that are almost always a good idea, no matter what task you give an agent. Limit 1: Do not share personal information. The agent should not share your address, your school name, your family members' names, or any other private details with anyone else. Limit 2: Do not spend money. Unless you and a trusted adult have specifically said it is okay, the agent should not buy, order, or pay for anything. Limit 3: Do not send or post anything without checking first. Messages, social posts, and emails should always be reviewed by you before they go out. Limit 4: Do not delete things permanently. Deleting something forever is hard to undo. A good rule is to move things to a trash folder first so you can check before they are gone for good. Limit 5: Do not contact strangers. The agent should only reach out to people you have already approved — not new people it finds on its own.
Flashcards — click each card to reveal the answer
Complete the sentence about setting rules for agents.
Think of limits like fences on a playground. A fence does not stop you from having fun — it actually makes the playground safer and more fun, because you know the boundary is there. Limits for an agent work the same way. They do not make the agent less helpful. They make the agent MORE helpful, because the agent can work freely inside the fence without accidentally causing problems outside it. When you set good limits, you can give the agent more freedom INSIDE those limits — because you know the boundaries are protecting you.
Limits do not make an agent less useful — they make it SAFER to use. A good fence around a playground lets kids play freely without worrying about running into traffic.
Match each limit to the problem it prevents.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
Why is it important to give an agent limits, not just instructions?
You are setting up an agent to help you plan a school project. Which of these is the best set of rules to add?
Write My Agent's Rule Book
- Imagine you are setting up an AI agent to help you with a school project or a hobby you love.
- On paper, write your agent's Rule Book with two sections:
- Section 1 — What the agent CAN do: list at least three tasks you want it to help with.
- Section 2 — What the agent CANNOT do: list at least four limits that keep you safe.
- Share your Rule Book with a trusted adult. Talk through: are there any limits you missed? Are there any limits that seem too strict?
- Keep your Rule Book somewhere safe — great agent bosses always know their rules!