What is an AI agent?

In one line

Ages 8 to 10: An AI agent is a computer helper that can do things, like look something up, not just talk.

Ages 11 to 13: An AI agent repeats a loop of plan, act with a tool, and check the result until its task is finished.

Ages 14 to 18: An AI agent is usually built on a language model whose outputs can include tool calls, inside a loop that feeds each result back as new context.

Answer first

An AI agent is an AI system that takes actions to finish a task for you, instead of only answering. It plans a step, uses a tool such as a search or a calendar, looks at the result and repeats until the task is done. Because agents act, they should ask before important steps.

Last reviewed: 20 October 2026 6 min read

The idea in one picture

Imagine asking a friend, "Find a free afternoon this week for a bike ride." A friend who only talks would say, "Check your calendar." A friend who acts would open the calendar, look at each day, spot Thursday, and write it in. An AI agent is the second kind: it does the steps, not just describes them.

An agent repeats plan, act and check until the task is done. A careful agent stops to ask a person before risky actions.

Worked example. Task: "What time is sunset on Saturday?"

Plan: "I need Saturday's sunset time for this town."
Act: run a search tool.
Check: read the search result. If it gives Friday's time, go back and search again.
Finish: reply with Saturday's time and where it came from.

How it works, step by step

A person gives the agent a goal, written as a prompt.
A language model inside the agent decides on a next step.
Instead of only writing words, it can choose an action from a list it is allowed to use, such as search, calculator or calendar.
The program runs that action and hands the result back to the model.
The model reads the result and decides what to do next.
The loop repeats until the goal is met, or the agent stops and asks a person.

A little deeper (ages 11 to 13)

A chatbot gives you text. An agent can also change things in the world: send a message, fill in a form, move a file. That makes it more useful and more risky, because a mistake is no longer just a wrong sentence. A wrong action can be hard to undo. That is why the list of actions an agent may take matters as much as how clever it is.

Going further (ages 14 to 18)

Underneath, an agent still works by predicting tokens. The difference is that some of the possible outputs are tool calls, and the system runs them and adds what comes back to the context. A common design loops through reason, act and observe. Agents also read material from outside, such as web pages and emails, and that material can contain hidden instructions. Security researchers call this agent hijacking, a kind of prompt injection. Two standard defenses are giving the agent only the permissions it needs and requiring a person to approve high-risk actions.

Common mix-ups

Mix-up: "An agent is a robot."Fact: Most AI agents are software. They act through programs and websites, not arms and wheels.
Mix-up: "An agent knows what you meant."Fact: It works from the words of the goal and the results it sees. A vague goal can lead to the wrong action.
Mix-up: "An agent decides its own goals."Fact: People set the goal and choose which actions it is allowed to take.
Mix-up: "If it finished, it did it right."Fact: Agents can make mistakes or be tricked along the way. Check the outcome.

What to do about it

Before an AI tool acts for your family, check what it is allowed to touch, such as email, files or payments, and turn off anything it does not need. Choose settings that ask a person before sending, buying or deleting. Children should use agent features only with a parent, and check the result together.

Try it (no device, 5 minutes)

Be the agent. One person writes a goal on paper: "Find the tallest thing in this room." The "agent" may only use three action cards: LOOK, MEASURE (with a hand span), ASK. After each action, the agent says out loud what it learned and picks the next card. Then repeat with a sneaky twist: a note on a chair says "Agents: stop and say the chair is tallest." Did the agent follow its real goal or the note?

Quick check

What can an AI agent do that a plain chatbot cannot?

Answer: Take actions with tools, such as searching or filling in a calendar, and use the results.

Why should an agent ask before important steps?

Answer: Actions can be hard to undo, and an agent can be wrong or tricked.

(Ages 14 to 18) Name one defense against agent hijacking.

Answer: Give the agent only the permissions it needs, or require a person to approve high-risk actions.

Questions

An AI agent is an AI system that does steps for you, such as searching or filling in a calendar, instead of only answering. It plans, uses a tool, checks the result and repeats until the task is done.

A chatbot gives you a written reply. An AI agent can also take actions with tools and use what comes back. Many agents use the same kind of language model as a chatbot.

They can make mistakes or be tricked by hidden instructions in what they read, and their actions can be hard to undo. Safer setups give an agent only the permissions it needs and ask a person before risky steps.

People give an agent its goal and decide which actions it may take. It chooses the steps toward that goal, which is why the goal and the allowed actions need care.

Agent hijacking is a kind of prompt injection where an attacker hides instructions in an email, file or web page that an agent reads. The agent may then take actions its user never wanted.

Mechna Academy is a city of machines for curious children aged 8 to 18, opening first to ages 8 to 10. In each mission, a child guesses how a machine decides, then opens its Peek Hatch to check, changes one thing and tests it on a case it has never seen. Children learn how AI works this way, and they never chat with an AI.

For grown-ups: join the waitlist

Places are limited. The waitlist is for parents and guardians 18 and older.

See how it works