Bots, chatbots, assistants, and agents

Companies use these four words loosely, and sometimes interchangeably, which is why they’re confusing. Here’s what each one usually means, and the question that matters more than the label.

---

Bot

Software that performs a task automatically.

A bot does not have to use AI. Plenty of bots follow a fixed rule and nothing more.

Example: a bot posts tomorrow’s weather forecast into a team chat every morning at 7. Nobody talks to it. It doesn’t understand anything. It runs a scheduled step.

---

Chatbot

Software you interact with through conversation. You type or say something, and it responds.

Chatbots come in very different kinds:

• Scripted chatbots pick from preset responses. The help window on a website that offers buttons like “Track my order” or “Talk to a person” is often one of these.
• Generative AI chatbots write a new response each time, based on what you typed. That’s what most people mean now when they say “chatbot.”
• Some conversational AI systems can also use tools and take actions, which is where the next two words come in.

---

AI assistant

An AI system designed to help you get things done, usually through conversation.

What separates an assistant from a basic chatbot is what’s been added around it. An assistant may have:

• Instructions from whoever set it up
• Reference material to draw on, like a company’s help articles
• A memory of earlier conversations, if that’s turned on
• Connections to other apps, such as your calendar or files
• Tools it can use, like searching the web or reading a document you upload

ChatGPT, Claude, Gemini, and Microsoft Copilot are all offered as AI assistants. Businesses also build custom assistants for their own customers or staff.

One naming trap: an assistant is a product. The AI underneath it, the trained system that produces the words, is a model, and the same name is sometimes used for both. The next page untangles that.

---

AI agent

An AI system that works toward a goal by choosing its own steps, and may use tools or take actions, instead of only writing a reply.

Ask a chatbot how to change a delivery address and it tells you how. Ask an agent to change it, and if it has the access, it may go and do it.

“Agent” means different things at different companies. Some use it for anything that can take an action. Some reserve it for systems that plan several steps on their own. Some use it as a marketing word for a chatbot. When you see the word, look for what the thing is actually allowed to do.

---

One example that shows the difference

Imagine an online store’s help window, and three questions a customer might ask.

“What time do you close?” A simple chatbot can handle this. It returns the answer it was given.

“Where is my order?” This needs a connected assistant. It looks up the order and explains the status in plain words.

“Please change the delivery address.” This needs an agentic system, one with both the permission and the tools to act. It may check that it’s really you, then make the change.

Same window. Very different capabilities behind it.

---

The question that matters more than the label

What information, permissions, and tools does it have?

A system can only look up your order if someone connected it to the orders. It can only change your address if someone gave it permission to. The label tells you less than the answer to that question.

That cuts both ways. Something called an “agent” may not be able to do much. Something called a “chatbot” may be connected to more than you’d expect.

---

Something you can try

Next time you open a help chat on a website, ask it:

What can you actually do for me here? Can you look up my account
or make changes, or can you only answer general questions?

Some will tell you plainly. If the answer is vague, assume it can do less than it sounds like, and ask for a person before you share anything sensitive.

---

Something important to know

The more a system can do, the more its mistakes can do. A chatbot that gives a wrong answer wastes your time. An agent that misunderstands you can take the wrong action. Before you let any AI act on your behalf, know what it’s allowed to change and whether you can undo it.

---

If it doesn’t match what you see

A product calls itself something different from what’s described here. Expected. These words aren’t regulated, and each company uses them its own way. Go back to the question above: what can it see, and what can it change?

An assistant says it did something, and nothing happened. It may have described the action instead of taking it, or a connection may have failed quietly. Check the real result (the order page, the calendar, the file) rather than trusting the message.