agents · Level 2
Agents vs chatbots: what's actually different
The one-question test, worked examples of each, and a five-question check you can run on any product that calls itself an AI agent.
Your next step
Start this course online
Register your start, then work through the lessons and exercises at your own pace. You can download the PDF separately. Where a scored exam is available, you will find it after the lessons.
Register or resume your course
Start with the essentials
The short answer
A chatbot replies to you inside a conversation: you ask, it answers, and you decide what happens next. An agent is given a goal and takes several actions on its own, choosing tools, checking results and continuing until the goal is met or it stops. The clearest test is simple: after one reply, who takes the next action, you or the system itself?
What you will learn
- You will be able to state, in one sentence, the real difference between a chatbot and an agent.
- You will be able to recognise worked examples of both, including a chatbot that looks clever and an agent that looks ordinary.
- You will be able to run a five-question diagnostic test against any product to work out which one it actually is.
- You will be able to see 'agentic' and 'AI agent' for what they often are, marketing language, and check the substance yourself.
- You will know why the distinction changes what you should expect and check before you rely on a tool.
Who it is for
Anyone who keeps hearing 'AI agent' in adverts, at work or in the news, and wants a plain answer to what actually makes something an agent rather than a chatbot. No coding needed.
Before you start
- Some experience of using an AI chatbot is helpful but not required. What is AI? and Your first prompts are useful preparation, not essential ones.
Learn in your browser · Chapter 01 of 06
The one question that actually separates them
Chatbot and agent get used almost interchangeably in adverts and app listings, but they describe two different things. One question cuts through the confusion, and this chapter shows you how to ask it.
A chatbot replies to you inside a conversation. You send a message, it sends back an answer, and you decide what happens next: copy the answer, act on it yourself, or ask a follow-up. However long the conversation runs, the pattern never changes: it answers, you decide.
An agent is given a goal rather than a single question. It decides its own steps towards that goal: which tool to use, what to do with the result, and whether to carry on or stop. It can work through several steps, one after another, without you doing the middle part yourself.
Scroll sideways to see every column.
| Chatbot | Agent | |
|---|---|---|
| What you ask for | How much does a week in Portugal usually cost in October? | Find three flight-and-hotel options for a week in Portugal in October under my budget, and shortlist the best |
| What happens next | It answers with a paragraph, you go and search for flights yourself | It searches, compares prices across options, and hands you a finished shortlist |
| Who did the middle work | You | The system |
Try it yourself · Activity 01
10 minSpot the pattern
Label each example as a chatbot or an agent, using the one-question test.
- A: you ask what the capital of Portugal is and get one answer.
- B: you ask an assistant to draft a holiday itinerary, read it, then book everything yourself.
- C: you tell a system to book the cheapest direct flight to Lisbon next Friday, and it searches fares, picks one and completes the booking.
- D: you ask a tool to find and email three hotels asking about a group rate, then tell you when they reply, and it does exactly that over the next day.
Which examples had you doing the middle work, and which had the system doing it? That gap is the whole difference.
Worked answer
A and B are chatbots: in both, you get a reply and you take the next step yourself, even though B involved a longer, more useful answer. C and D are agents: in both, the system carries out several steps itself (searching, booking, emailing, waiting for a reply) without you doing them.
Learn in your browser · Chapter 02 of 06
Chatbots in the wild: clever, but still one reply at a time
A chatbot can be genuinely impressive and still be a chatbot. The tell is not how good the answer is, it is what happens after the answer.
Take a shop's website chat widget. It answers questions about sizing, returns and delivery, sometimes by checking the shop's own product list before it replies. That lookup can make the answer accurate and specific, but it is still one action taken to answer one question. You read the reply, then you decide whether to buy, ask again or leave. The chatbot never goes off and does the buying, returning or emailing for you.
Take a general AI assistant used to draft an email. You ask for a draft, it writes one, you read it, ask for changes, it rewrites, and you go back and forth several times. That can feel like a working relationship rather than a single question and answer, but the pattern underneath has not moved: at every turn, the assistant replies and stops, and you decide what happens with the draft, including whether it actually gets sent.
Scroll sideways to see every column.
| Trait | What it looks like in a chatbot |
|---|---|
| Who replies | It answers once per message you send |
| Memory | Usually just the current conversation |
| Can it act outside the chat | Only if you ask it to, one action at a time, inside that turn |
| Who decides the next step | You, every time |
Try it yourself · Activity 02
10 minWas that actually a chatbot?
Pick an AI tool or app you have used recently.
- Write down the last thing you asked it to do.
- Write down exactly what it did in response.
- Ask: did it stop and wait for your next message, or did it go off and do something else on its own?
- Ask: across the whole conversation, who decided what happened after each reply?
If you were the one deciding after every single reply, however long the conversation ran, you were using a chatbot, however good it felt to talk to.
Learn in your browser · Chapter 03 of 06
Agents in the wild: several steps, chosen by the system
The examples below could not happen inside a single reply. That is exactly what makes them agents rather than chatbots.
Take a coding helper given one instruction: fix the failing test. It reads the error message, opens the relevant file, edits the code, reruns the tests, sees they still fail, tries a different fix, and reruns them again, repeating until they pass or it runs out of attempts. Nobody approved each individual edit as it happened. The system planned, acted, checked the result and decided what to do next, several times over, on its own.
Take a research-and-booking task: find three suppliers, compare their quotes, and set up a call with the cheapest one. An agent given that goal searches for suppliers, opens their pages, emails each one for a quote, waits, reads the replies as they arrive, compares the numbers, and books a time in a calendar with the cheapest supplier, adjusting its plan if a reply is slow or a supplier turns out to be unsuitable. None of that happens in one exchange, and none of it needed you to carry out the middle steps.
Scroll sideways to see every column.
| Trait | What it looks like in an agent |
|---|---|
| Who acts | The system itself, using tools it chooses |
| How many steps | As many as the goal needs, decided as it goes |
| Can it work while you are away | Yes, that is the point of giving it a goal rather than a single instruction |
| Who decides the next step | The system, based on what happened last |
Try it yourself · Activity 03
15 minDesign a one-step agent test
Think of a product that describes itself as an AI agent.
- Give it, or imagine giving it, a small task with at least three real steps, such as find, compare and shortlist three of something.
- Watch, or predict, whether it does the searching and comparing itself, or just tells you how to do it.
- Count how many steps happened before it needed you again.
- Decide honestly: did the system do the middle work, or did you?
A product that needs you to carry out every middle step, however useful its single replies are, is not really acting as an agent yet, whatever it is called.
Learn in your browser · Chapter 04 of 06
The five-question test: work out what you are actually using
You do not need to trust a product's own label. Five short questions tell you what a tool is actually doing, whatever the marketing calls it.
- When you give it a task, does it wait for your next message, or does it go off and do something (search, click, edit, send) on its own?
- If it does act, does it look at what came back and decide the next step itself, or does it just report back and stop?
- Can it complete the whole task, start to finish, without you carrying out the middle steps yourself?
- Does it choose which tool to use for this task, rather than you picking one tool for it each time?
- If you closed the app and came back later, could it have finished the task without you there?
Scroll sideways to see every column.
| Question | Product A: a support chat that checks stock before answering | Product B: given 'find and shortlist three suppliers' |
|---|---|---|
| Waits for your next message? | Yes, after each stock check | No, keeps going by itself |
| Decides the next step itself? | No, it just answers | Yes, it searches, reads and compares |
| Completes the whole task alone? | No, you still have to buy or ask further | Yes, hands you a finished shortlist |
| Chooses its own tool? | No, it always uses the same stock lookup | Yes, it picks search, page-reading and comparison as needed |
| Could finish it while you're away? | No | Yes |
Try it yourself · Activity 04
15 minTest something you actually use
Pick one real AI tool or app you use, at work or elsewhere.
- Write down the five questions above, with space to answer each one.
- Give the tool a small, real, low-stakes task, or recall the last time you used it for one.
- Answer each question honestly, based on what it actually did, not what its own description says.
- Decide: chatbot, agent, or a hybrid with one tool bolted on.
Did the tool's own name or description match what you actually found once you tested it?
Worked answer
Example: a note-taking assistant that summarises a meeting recording. Waits for your next message: yes, it produces one summary and stops. Decides the next step itself: no. Completes the whole task alone: only partly, the summary is finished but filing or acting on it is still yours. Chooses its own tool: no, it transcribes then summarises the same way every time. Could finish it while you're away: no, you still have to read and act on the summary. Verdict: a chatbot, a genuinely useful one, but a chatbot.
Learn in your browser · Chapter 05 of 06
Why the label alone cannot be trusted
Most of this confusion comes from marketing, not technology. Knowing why makes the five-question test easier to trust.
Agentic is used loosely across the industry to describe anything from a chatbot with a single tool attached to a system that plans and acts through many steps unsupervised. Both get called agentic. Neither use is exactly wrong, since agentic describes a spectrum rather than a strict category, but it means the word on its own tells you very little about what a specific product actually does.
It is not just a naming argument. The two patterns carry genuinely different risks. With a chatbot, you see and approve every step, because you are the one taking each next action. With an agent, several steps can happen before you see anything at all, which means an early mistake can be built on by later steps before anyone notices. That does not make agents unsafe to use, but it does mean the checking has to happen differently: reviewing the result at the end is not the same as approving each step as it happens.
Scroll sideways to see every column.
| Hybrid example 1 | Hybrid example 2 | |
|---|---|---|
| What it does | A chatbot drafts an email and, once you approve it, sends it for you | An assistant plans several steps, and once you approve the plan, carries all of them out unsupervised |
| Leans towards | Chatbot, since you approve every send | Agent, since the execution afterwards runs without you |
None of this is about which pattern is better. A chatbot is often exactly right: predictable, and you check every step because you are doing every step. An agent earns its extra risk only when the task genuinely needs several steps you would rather not do by hand, and you can tolerate checking the outcome rather than each individual action along the way.
Learn in your browser · Chapter 06 of 06
What you can do now
You can now tell a chatbot from an agent without relying on a product's own label, and check any tool for yourself.
You can now state the one-question test, recognise worked examples of both patterns even when they look more advanced than they are, and run the five-question test against any real product you use. You have also seen why 'agentic' on its own does not tell you much, and why the two patterns carry different risks worth checking differently.
This is free educational content published by Mickai LTD, not professional, legal, financial or compliance advice, and Mickai accepts no liability for how you use it.
Being able to say exactly what a system does, rather than trusting what it is called, is the same principle behind Mickai's approach to trustworthy AI: understand the real behaviour of a tool before you rely on it, not just its label.
https://trust-agent.ai/exam/agents-vs-chatbots-whats-differentKeep learning
The complete workbook
This workbook untangles one of the most common mix-ups in AI conversation: chatbot and agent are not the same thing, and the difference is not marketing spin. You will see worked examples of each, including cases that look more advanced than they are, learn the one question that tells them apart, and run a five-question test against any real product to work out what you are actually using.
- 01The one question that actually separates themRead here · 1 exercise
Chatbot and agent get used almost interchangeably in adverts and app listings, but they describe two different things. One question cuts through the confusion, and this chapter shows you how to ask it.
- 02Chatbots in the wild: clever, but still one reply at a timeRead here · 1 exercise
A chatbot can be genuinely impressive and still be a chatbot. The tell is not how good the answer is, it is what happens after the answer.
- 03Agents in the wild: several steps, chosen by the systemRead here · 1 exercise
The examples below could not happen inside a single reply. That is exactly what makes them agents rather than chatbots.
- 04The five-question test: work out what you are actually usingRead here · 1 exercise
You do not need to trust a product's own label. Five short questions tell you what a tool is actually doing, whatever the marketing calls it.
- 05Why the label alone cannot be trustedRead here · Reading
Most of this confusion comes from marketing, not technology. Knowing why makes the five-question test easier to trust.
- 06What you can do nowRead here · Reading
You can now tell a chatbot from an agent without relying on a product's own label, and check any tool for yourself.
Also inside: a 8-point checklist, a glossary of 10 terms and 10 questions and answers to test yourself. 4 hands-on exercises, each with a worked answer at the back where the workbook gives one.
The PDF is free. We ask for your name and email before downloading. Read the privacy notice at the download form. Free to download and use for personal learning, study groups and inside your own team. Please do not resell the workbooks or republish them as your own. Link people to trust-agent.ai instead.
Get your free PDF workbook
Enter your name and email, then download the complete workbook. This form also works without JavaScript.
Test yourself
Questions and answers
What is the one question that actually separates a chatbot from an agent?
After a reply, who takes the next action? If you do, whatever you go on to do with the answer, it is a chatbot. If the system itself goes on to act, such as searching, editing or sending, without you doing that step, it is an agent.
Can a chatbot use a tool, such as checking a fact, and still be a chatbot?
Yes. A single lookup used to answer one question is still one action taken to produce one reply. Carrying out a task with several of its own steps is a different thing from fetching one fact before answering.
What's an example of an AI tool that looks advanced but is actually still just a chatbot?
A long, back-and-forth conversation drafting and redrafting an email. However many turns it takes, the assistant replies and stops at each one, and you decide what happens with the draft, including whether it is sent.
How can you tell if an AI product is a genuine agent rather than a chatbot with extra steps bolted on?
A genuine agent is given a goal once, then plans, acts, checks the result and decides what to do next by itself, repeatedly, without a person doing the middle steps in between. A chatbot, by contrast, replies once and waits for you to decide what happens next.
What are the five questions in the diagnostic test?
Does it wait for your next message or act on its own? Does it decide the next step itself? Can it finish the whole task without you doing the middle part? Does it choose its own tool? Could it finish the task while you are away?
What does a real mix of yes and no answers to the five-question test usually suggest?
It often points to a hybrid product, such as a chatbot with one tool bolted on. It can be genuinely useful, but it is not the same as a system that completes a task on its own from start to finish.
Why can't you trust the word 'agent' on a product page by itself?
The industry uses 'agentic' loosely, covering everything from a chatbot with one attached tool to a system that plans and acts through many unsupervised steps. The label does not tell you which one you actually have.
What is a hybrid product, and what is an example?
A product with traits of both patterns, such as a chatbot that drafts an email and, once you approve it, sends it for you. It leans towards chatbot when you approve every step, and towards agent once execution afterwards runs unsupervised.
Why does it matter, practically, whether a tool is a chatbot or an agent?
The two carry different risks. With a chatbot you see and approve every step. With an agent, several steps can happen before you see anything, so an early mistake can be built on before anyone notices, which changes how you need to check the result.
What should you do if a product does not say plainly whether it acts on its own?
Treat that silence as useful information in itself, and run the five-question test yourself on a small, real task rather than assuming either answer from the marketing.
Write as you go
My notes
Use this scratchpad for your own notes, working or answers as you read. Notes are saved in this browser on this device. Downloading makes a local copy; it does not send your notes to Mickai.
Consolidate your learning
Review your work
Work through the exercises and questions, then compare your reasoning with the worked answers. A scored exam and verifiable certificate are not yet available for this course.
Review the questionsLearn the language
Key terms
- Chatbot
- A system that replies to you inside a conversation, one message at a time, leaving you to decide and carry out whatever happens next.
- Agent
- A system given a goal that chooses its own steps towards it, using tools, checking results and continuing until the goal is met or it stops, without you carrying out the middle steps.
- Agentic
- A loosely used industry term for AI that works towards goals through actions and steps rather than a single reply. It describes a spectrum, not a strict category.
- Tool
- An action a system can take, such as searching, reading a file or sending a message. A chatbot might use one tool per reply; an agent chooses which tools to use as it works through a task.
- Tool call
- One instance of a system using a tool, such as running one search or sending one message, as part of answering you or carrying out a task.
- Multi-step task
- A task that needs several actions carried out in sequence, with each step depending on what happened before it, rather than one action answering one request.
6 of the workbook's 10 terms. The complete glossary is in the workbook.
Follow the evidence
Sources and checks
Facts last checked: .
These workbooks use AI assistance. See how the workbooks are made.
Created by Mickarle Wagstaff-Irons - Micky Irons with the Mickai team. Published by Mickai LTD. Last updated 29 September 2026.
NextKeep going
Where to go next
Recommended for you
What is an AI agent? Agentic AI explained
A plain-English guide to AI agents: how they differ from chatbots and workflows, the parts and the loop, levels of autonomy, the main risks and how to keep a human in control.
Recommended for you
Build your first chatbot
Build a chatbot that behaves: model access, system prompt, history and context limits, guardrails, tone, a test set, logging and cost, on a no-code path or a code path.
Recommended for you
Deploy AI agents without losing control
Plan least-privilege access, prohibitions, approval thresholds and a checked log for AI agents. Exercises distinguish written plans from implemented controls.