Chatbot or agent? What an AI agent actually does

A chatbot answers your questions. An agent can also do a job for you, step by step, and you check the result.

Explanation

An answer, or an errand?

A chatbot answers in a conversation. You ask "What's a revision timetable?" and it explains. You can ask follow-up questions.

An agent can also take steps towards a goal. You might say "Here are my test dates and topics. Draft a revision timetable and ask me which topic I find hardest first." A capable agent can read what you gave it, ask its question, organise the topics and hand you a draft to check.

Think of the difference between someone who answers questions and a helper you can send on an errand. What matters is what the helper can actually do, and what you've said it may do.

What makes something an agent?

An agent can usually do some or all of these:

• Use tools: search the web, read a page, or use a service's own tools. Apex has tools like this.
• Take several steps: read, plan, do the work, then check it.
• Remember context: keep track of what you've said in the conversation. Some apps can also remember things between chats.
• Work when you hand it a task: start a job when you ask, and in some apps run a task later or on a schedule.

Chat apps differ here. Many offer some of these features but not others, and free and paid plans often differ too. You can't tell from the chat box alone. Lesson 2 shows what's known about each app.

How does the work start?

Signing up gives you access to an app. It doesn't give the AI a job. Work starts when you hand it a clear task:

Help me plan revision for three topics: fractions, percentages and ratios.
I have 20 minutes each evening for three evenings.
Ask me which topic I find hardest before you plan.
Just write the plan here. Don't use any of my accounts.

A good agent asks its question first, then drafts the plan. Your job is to check that the plan really fits your week.

For a job on a website, such as Apex, the app needs a way to open that site or use its tools. Saying "you're allowed to" can't add a tool the app doesn't have. Some apps can't connect to Apex at all, and lesson 2 tells you what to look for.

You stay in charge

Agents make mistakes. They can misunderstand you, use a poor source, or say they've finished when they haven't. So:

• Start with small tasks and drafts.
• Say what it may do and where it must stop. "Show me the draft first" is a great limit.
• Check the result before you rely on it.

Want to go further? Apex's free lessons Chat assistants vs agents vs integrations and Agents and tool use explain this in more depth.

Try it

Write down one question you'd like answered and one small job you'd like done. For the job, add the result you want and one limit. Then ask your app, "Can you open a web page or use a tool for me?" Check its answer against the app's own help pages, because AI can be wrong about itself too.

Quiz

Q1. Which request is a task for an agent rather than a simple question?

• "What does the word 'ratio' mean?"
• "Read these topics, ask me which is hardest, then draft a three-evening revision plan."
• "Can you explain percentages in a friendlier way?"

Q2. You've just created an account with an AI app. What starts useful work?

• Waiting for the app to suggest a project for you.
• Giving it a clear task with the result you want and a limit.
• Assuming it has already linked itself to your other accounts.

Q3. Your agent says, "Done! Everything's finished." What should you do next?

• Start the next task straight away, because it said it finished.
• Trust it, because it sounded confident.
• Look at the result and check it against what you asked for.

Sources

• Apex lesson 17: Chat assistants vs agents vs integrations
• Apex lesson 27: Agents and tool use
• Google Cloud: What are AI agents?, checked 18 Oct 2026