"Chatbot", "AI assistant" and "AI agent" are often used as if they meant the same thing, and vendors stretch all three. For people at work the difference matters, because it decides what the software can actually change, and how much checking it needs.
The short answer: a chatbot answers. An assistant helps you do your task, with you making each move. An agent is given a goal and takes actions itself, such as sending, updating or buying, inside real systems. The more an AI system can act on its own, the more care it needs: clear limits on what it may touch, human approval for big actions, and a named person who answers for the result.
Three kinds of AI help, side by side
| Chatbot | AI assistant | AI agent | |
|---|---|---|---|
| What it does | Answers questions in a conversation | Drafts, summarizes and suggests inside your work | Plans and carries out multi-step tasks toward a goal |
| Who makes each move | You read the answer and decide | You accept, edit or reject each suggestion | The agent, within the limits it is given |
| Can it change things? | No, it only replies | Usually only when you click to apply | Yes: it can send, update, file or order through connected tools |
| Typical example at work | Answering staff questions about the leave policy | Drafting a reply to a customer email for you to send | Handling a refund request end to end, from checking the order to issuing the refund |
| Main risk | A wrong answer someone relies on | A wrong draft that slips through | A wrong action taken in a real system |
| Human role | Read critically | Review and own the final version | Set the goal and limits, approve high-impact steps, check results |
These are working categories, not legal or technical standards; the same product can behave as a chatbot in one setting and an agent in another, depending on what it is connected to.
What makes something an agent
The key difference is action. The ILO describes the models behind these tools as gaining agentic capabilities that allow them to execute multi-step tasks autonomously, interact with software environments and make decisions based on contextual inputs (ILO Working Paper 140, 2025). In practice an AI agent usually has:
- A goal rather than a single question.
- Tools: access to email, files, databases, calendars or business systems.
- Steps: it decides what to do next based on what it finds.
- Permissions: what it is allowed to read and change.
A chatbot connected to the company handbook is still a chatbot. Give it the ability to change someone's leave balance and it has become an agent.
Why the difference matters for risk
The security project OWASP describes "excessive agency" as a top risk for AI applications. It notes that such systems are often granted agency, the ability to call functions or interface with other systems to undertake actions, and that problems come from excessive functionality, excessive permissions and excessive autonomy. Its recommended controls include human-in-the-loop control to require a human to approve high-impact actions, and limiting tools to the minimum functionality and permissions needed (OWASP LLM06:2025 Excessive Agency).
All three kinds of tool can be wrong; NIST calls confidently stated false output confabulation (NIST AI 600-1). The difference is what happens next. A wrong chatbot answer misleads a reader. A wrong agent action has already happened. See AI agent security.
Telling people they are dealing with AI
In the EU, Article 50 of the AI Act requires providers to ensure that AI systems intended to interact directly with people are designed so that the people concerned are informed that they are interacting with an AI system, with some exceptions, such as where it is obvious (EU AI Act, Article 50). That applies whether the system is a simple chatbot or a customer-facing agent.
What this means for the people working with them
- With a chatbot, the skill is asking well and reading critically. See checking AI output.
- With an assistant, the skill is editing and owning what goes out under your name.
- With an agent, the skills are management ones: writing a clear brief, setting limits, reviewing logs and deciding which steps need approval. See AI agents as coworkers and human in the loop and accountability.
That shift, from using a tool to supervising one, is what we call digital collar work. For practical cases, see AI agent examples at work, and for definitions of related terms, our glossary.
Questions and answers
What is the difference between an AI agent and a chatbot?
A chatbot answers questions in a conversation. An AI agent is given a goal and takes actions itself, such as sending messages or updating records, through connected tools.
Is an AI assistant the same as an AI agent?
Not usually. An assistant drafts and suggests while you make each move; an agent carries out steps on its own within the limits it is given.
Can a chatbot become an agent?
Yes. When a chatbot is connected to tools that let it change things, such as records, orders or emails, it starts acting as an agent and needs the controls that go with that.
Why are AI agents riskier than chatbots?
Because they act. OWASP lists excessive agency, meaning too much functionality, permission or autonomy, as a top risk and recommends human approval for high-impact actions.
Do companies have to say when I am talking to AI?
In the EU, Article 50 of the AI Act generally requires AI systems that interact directly with people to be designed so people are informed, unless it is obvious.