At a glance
| Chatbot | Assistant | Agent | |
|---|---|---|---|
| Main job | Hold a conversation, answer questions | Help with tasks you request | Achieve a goal across multiple steps |
| Who decides the steps? | You, turn by turn | You, with the system executing | The system, within limits you set |
| Tools / actions | Usually none or few | Some (search, files, calendar), often one at a time | Many; used repeatedly in a loop |
| Autonomy | Low | Low to medium | Medium to high |
| State across turns | Conversation history | History plus some preferences | History, task state, memory |
| Main risk | Wrong or off-topic answers | Wrong answers; mis-used tool | Compounding errors; unintended actions |
Marketing uses these words interchangeably, so judge a product by what it does rather than what it is called.
What is each?
Chatbot. Software you talk to. Older chatbots followed scripts and decision trees; modern ones are powered by LLMs. Their output is words.
Assistant. A conversational system that also helps get things done on your instruction: drafting an email, searching a document, creating a calendar event. Typically you remain in the driver's seat — it does what you ask, often with confirmation.
Agent. A system that takes a goal and works out and carries out the steps itself, calling tools, checking results and adapting (What are AI Agents?). You describe the destination, not the route.
Key differences
The difference is about who controls the loop.
- In a chatbot or assistant, control returns to you after each response.
- In an agent, the model keeps going — plan, act, observe, repeat — until it believes it is done or a limit is reached.
Consequences:
- Reliability: each autonomous step can go wrong, and mistakes compound.
- Cost and latency: agents make many model and tool calls.
- Oversight: agents need guardrails — step limits, scoped permissions, approvals, logs.
- Security: more tools plus more autonomy increases exposure to What is Prompt Injection?.
The same product can sit at different points: an assistant that offers "go and do this for me" is switching into agent mode.
When to choose which
- Chatbot — FAQs, explanations, guidance where no action is needed.
- Assistant — drafting, summarising, lookups, structured help where you review and decide.
- Agent — multi-step tasks where the steps depend on what is discovered (debugging, research, investigation) and the actions are safe, reversible or approved.
- Neither — if the steps are fixed, use an agentic workflow or plain code. Autonomy you do not need is risk you do not need.
Example
"Why did last month's revenue dip?"
- Chatbot: explains common causes of revenue dips.
- Assistant: queries the revenue table when you tell it which one and shows you the numbers.
- Agent: decides to compare regions, notices one region dropped, pulls its orders, finds a payment-provider outage in the logs, and reports a sourced explanation — having chosen those steps itself, read-only.
Common mistakes
- Calling every LLM app an "agent".
- Choosing an agent because it sounds advanced.
- Giving an agent write access before it has earned trust through testing (How to Evaluate AI Systems and Outputs).
- Forgetting that a chatbot with retrieval (What is RAG (Retrieval-Augmented Generation)?) is still not an agent — it is still a single-pass answer.