A chatbot answers questions in a chat window. An AI agent takes actions in your other tools: it can read and send email, update records, book appointments or move files. The same AI model can power both. The difference is access, and access is what decides how much can go wrong.
Think of it as three levels of access:
| Tool | What it can do | What can go wrong |
|---|---|---|
| Chatbot | Answers questions in a chat window | A wrong or made-up answer |
| Assistant | Drafts things for a person to use (emails, summaries, posts) | A person copies a wrong draft |
| Agent | Acts in your tools on its own (sends, updates, books, pays) | A wrong action nobody reviewed |
Vendors use these words loosely, so don’t go by the label. Ask what the tool is allowed to do once it’s connected. If the answer includes sending, changing, deleting or paying, treat it as an agent and set limits before it goes live.