From chatbots to agents

In this lesson you'll

• Explain, in one or two sentences, the difference between a chatbot and an AI agent.
• Name the three things that make something an agent: a goal, a plan of steps, and tools it can use to act.
• Recognize where the Ask, Partner and Assign ways of working fit, and why Assign is the focus of this course.
• Spot the new responsibilities that come with letting AI take actions for you.

Why this course, and why now

The Enterprise Technology Association (ETA) created this course for Level 1: Foundations of the ETA Certified AI Empowered Professional certification.

Most people meet AI through a chatbot: you type a question, it types an answer. That is useful, but you still do all the work of acting on the answer. You copy the text into an email. You open the website. You update the spreadsheet.

A newer kind of AI tool goes further. It can open a browser, read pages, fill in forms, work inside your documents, or run a series of steps while you do something else. These tools are called AI agents. Agent features are showing up in many common assistants and office apps, so many working adults will meet one soon, whether or not they go looking for it.

This course helps you use agents with confidence and good judgment. You do not need any technical background. If you have asked a chatbot a question before, you are ready.

Chatbot vs. agent in plain words

A chatbot (also called an AI assistant) is a program you talk to in everyday language. It answers questions, explains ideas and drafts text. It responds to you, one message at a time, and then it waits.

An AI agent is an AI system that is given a goal, makes a plan of steps to reach it, and uses tools to take actions, checking its progress as it goes. A tool here means any ability that lets the AI do something beyond writing a reply: searching the web, clicking buttons on a website, reading a file, editing a spreadsheet, or sending a message through a connected app.

The simplest way to remember it:

A chatbot answers. An agent acts.

A workplace example: the same job, two ways

Priya coordinates training at a regional hospital system. She needs to find three upcoming, low-cost workshops on patient privacy for her staff.

With a chatbot, Priya asks, "What are good patient privacy workshops?" The chatbot gives general suggestions. Priya then searches each one herself, checks the dates and prices, and builds a short list in a document.

With an agent, Priya writes: "Find three patient privacy workshops for hospital administrative staff held in the next 90 days, under $300 per person, online or within 50 miles of Columbus, Ohio. Put the name, date, price and link in a table. Do not register for anything." The agent searches, opens the event pages, checks dates and prices, and builds the table. Priya reviews the table and clicks the links to confirm before sharing it.

Same goal. The difference is who does the steps.

The three parts of any agent

Whatever the brand or product, you can recognize an agent by three parts:

1. A goal. You describe the outcome you want, not every click.
2. A plan. The agent breaks the goal into steps, such as "search, open the top results, compare, write the table." Many agents show you this plan, and some let you edit it before they start.
3. Tools to act. The agent uses its tools to carry out each step, looks at the result, and decides what to do next.

That loop of plan, act, check and adjust is what lets an agent handle multi-step work. It is also why agents can go wrong in new ways: a small mistake in step two can carry through to step ten.

Ask, Partner, Assign: three ways to work with AI

ETA describes three ways people work with AI:

• Ask: the AI answers or drafts on request. ("Summarize this policy in five bullets.")
• Partner: you and the AI go back and forth to think or create together. ("Help me plan this workshop; push back on my ideas.")
• Assign: the AI carries out multi-step work for you. This is where agents come in. ("Collect the registration numbers from these five vendor pages and put them in a sheet.")

You will use all three. Assign is not "better"; it is a different tool for a different job. Many tasks are still best done with a quick Ask or a thoughtful Partner session.

What changes when AI can act

When an AI only writes text, the worst outcome is usually a wrong or awkward answer that you catch before using it. When an AI can act, mistakes can leave the chat window. An agent could send an email to the wrong person, delete a file, or submit a form with the wrong details.

That does not mean agents are dangerous to try. It means a few new habits matter:

• Be clear about the goal and the limits. The agent will fill gaps with its own guesses.
• Keep approval for actions that matter, such as sending, buying or deleting.
• Check the work and the path. Review what it produced and what it did to get there.
• Give only the access it needs. Fewer connected accounts means fewer ways things can go wrong.

You will practice every one of these in Module 2.

Try it

You can see the "plan, then act" idea even in a regular chatbot. Open your organization's approved AI tool (for example ChatGPT, Claude, Microsoft Copilot or Gemini) and paste this prompt. It asks the AI to plan a task as an agent would, without doing anything yet.

I want to understand how an AI agent would approach a task.
Task: "Find three free webinars on workplace safety in the next 60 days and list the name, date, host and link."
Please do NOT do the task. Instead:
1. List the steps an AI agent would take, in order.
2. Name the tool it would need for each step (for example web search, open a web page, write a table).
3. Mark any step where a human should check or approve before the agent continues.
4. Name two things that could go wrong.

Read the plan. Notice how many steps sit behind a "simple" request, and which steps you would want to check yourself.

Key takeaways

• A chatbot answers; an agent plans steps and uses tools to act toward a goal.
• Every agent has three parts: a goal, a plan and tools.
• ETA's three ways of working are Ask, Partner and Assign; agents are the Assign mode.
• Because agents act, you are responsible for clear goals, approvals, checking the work and limiting access.