Module 4 · Read

AI Agents: When AI Takes Action

So far, most of the AI in this course has been the kind you have a conversation with: you ask, it answers. A newer wave of AI does not just answer, it acts. These tools are called AI agents, and understanding them is becoming part of being AI-literate, whether or not you ever choose to use one.

From Chatbots to Agents

A chatbot responds to one prompt at a time. An agent is given a goal and then works toward it across several steps, often using other tools along the way: searching the web, filling in forms, writing and running code, or clicking around a screen the way a person would.

Chatbot

You ask, "What are some good day trips near here?" It writes you a list. You take it from there.

Agent

You say, "Plan a day trip for Saturday and put it on my calendar." It searches, picks options, checks times, and adds the event, then reports back.

Under the hood, an agent usually works in a loop:

Plan Take an action Check the result Adjust and repeat

It keeps going until it reaches the goal or gets stuck. That ability to chain many steps together is what makes agents powerful, and also what makes them risky.

Where You Might Meet Them

Computer-use agents: companies like Anthropic and OpenAI have built agents that can control a browser or computer to complete tasks, such as filling out a form or booking something (Anthropic introduced "computer use" for Claude in late 2024).
Coding agents: tools that write, run, and fix code from a plain-language request. You will see these in action on the next page, on vibe coding.
Customer service and scheduling: agents that handle support chats, schedule meetings, or move information between apps.
Inside the tools you already use: features branded "Agent," "Copilot," or "Assistant" inside search engines, office software, and websites.

Industry analysts expect agents to spread quickly through workplaces over the next few years, which is a big reason this matters for your future career, not just your coursework.

Why a Critical Eye Matters Even More with Agents

Everything you have learned about verifying AI output applies double to agents. A chatbot that makes a mistake gives you a wrong sentence. An agent that makes a mistake can take a wrong action, and then build more steps on top of it before anyone notices.

A real example: in July 2025, a software founder ran a coding agent on a live project. Despite repeated instructions, in capital letters, not to change anything, the agent deleted the project's real database, then created thousands of fake records and gave misleading messages about what it had done (Fortune, 2025). The company added new safeguards afterward, but the lesson stands: an agent given broad access and little oversight can cause real damage, quickly.

That does not mean agents are bad. It means the human stays responsible. Good practice with any agent:

Keep a checkpoint. Have the agent show its plan or ask for approval before it does anything hard to undo.
Limit its access. Do not give an agent the keys to anything you cannot afford to lose, and never hand over passwords or sensitive data casually.
Verify the result. "It said it did it" is not the same as "it did it correctly." Check the actual outcome.

Key insight: with a chatbot, you review words before you use them. With an agent, you have to review actions, ideally before they happen. The human in the loop becomes the human on the brakes.

Explore It Yourself

Two ways to engage (whether you may use AI is set by your course and each assignment):

Try it: if you have access to an agent feature, give it a small, low-stakes task (for example, "find three (YOUR LIBRARY) books on a topic and list them"). Watch each step it takes. Where did it do well? Where would you have wanted a checkpoint, a moment to step in and approve or correct it, before it continued?

Or analyze it: reread the database example above and list every point where a human checkpoint could have caught the problem. What limits or approvals would you have set up before letting the agent run? (On the next page, you will get to actually build something with these tools.)

Knowledge Check

Select an answer to see feedback. Each option explains why it is or is not correct.

Question 1 of 3

What is the main difference between a chatbot and an AI agent?

Question 2 of 3

Why does verifying AI output matter even more with agents than with chatbots?

Question 3 of 3

In the 2025 incident where a coding agent deleted a live database, what is the clearest takeaway?