Course 9, lesson 84 of 100, Ages 14+
AI agents
Models that take steps
Like I’m 5
A chatbot is like asking a friend a question. An agent is like asking a friend to go and do a job for you, step by step.
The big idea
An agent is an AI that can take actions, not just reply. Given a goal, it plans steps, calls tools like search, calculators or code, looks at the results and decides what to do next.
Agents can book meetings, research topics or fix code. Because they act in the world, they need guardrails: limited permissions, confirmation before risky actions, and logs so people can see what happened.
Examples
- Research agent: Searches several sites, compares facts, writes a summary.
- Coding agent: Edits files, runs tests, fixes failures.
- Guardrails: Asks 'Send this email?' before sending.
How it works
- A chatbot answers. An agent can also take steps: search the web, use apps, or write and run code.
- It makes a plan, does one step, checks the result, then decides what to do next.
- Agents need clear goals and limits, and a person should approve anything important.
Check your understanding
- What makes an agent different from a chatbot?
- Options: It can take steps and use tools; It has a robot body; It never makes mistakes.
Answer: It can take steps and use tools. Agents use tools and work in steps towards a goal. - Why should agents ask before risky actions?
- Options: Mistakes in the real world can be costly; It makes them slower for fun; Agents can't act.
Answer: Mistakes in the real world can be costly. Confirmation keeps people in control of important actions.
Remember
Agents plan, use tools and check their work.
Talk about it
What chore would you trust an AI agent with, and what wouldn’t you?
Go deeper
Agents run a loop: plan, call a tool, observe the result, repeat. Reliability drops as tasks get longer, so good agent design adds permissions, logs, human approval for risky actions, and testing on realistic tasks.