Chatbot, AI assistant, AI agent you’ve probably seen all three terms used in the same headline, sometimes describing the same product. They’re not interchangeable, though marketing copy often treats them that way. The real difference comes down to one question: how much is the software allowed to do without you approving each step?
The Quick Answer
- A chatbot answers. It replies to what you type, based on scripts or a trained model, and stops there.
- An AI assistant helps you do things, one request at a time: set a timer, answer a question, open an app, but waits for your next instruction.
- An AI agent takes a goal and carries out the steps to reach it on its own, checking in only when it needs to.
Each sits one rung higher on a ladder of autonomy than the one before it.
Full Comparison
| Chatbot | AI Assistant | AI Agent | |
|---|---|---|---|
| What it does | Answers a question or follows a script | Completes single requests on command | Works toward a goal across multiple steps |
| Starts action on its own | No | No — waits for you to ask | Often, once given a goal |
| Remembers context mid-task | Rarely, beyond the current message | Limited, within one request | Tracks progress through the whole task |
| Uses outside tools/apps | Rarely | Sometimes (e.g., opens an app) | Regularly — search, bookings, forms, other software |
| Everyday example | Website “chat with us” support bubble | Siri, Alexa, Google Assistant | A tool that plans and books your trip end to end |
Real Examples You Already Use
Chatbot: The support bubble on a retailer’s website that answers “where’s my order?” from a script, and hands you off to a human the moment your question doesn’t match a pattern.
AI assistant: Ask Siri to set a 20-minute timer, and it does exactly that one clear instruction, one action, done. It’s helpful and fast, but it’s not deciding what you need next; you’re still driving.
AI agent: Tell a travel-planning agent “book me a weekend in Lisbon under £400,” and it researches flights, compares hotels, checks they fit your budget together, and hands you a finished plan a string of decisions made on your behalf toward one goal, not a single command carried out.
The line blurs in real products. Siri increasingly picks up agent-like features, and many “AI agents” are really assistants with better branding but the underlying test still works: does it complete one instruction, or does it pursue a goal across several steps without you directing each one?
Where “Agentic AI” and “Generative AI” Fit Into This
These two get confused with the assistant/agent comparison too, though they’re answering a different question — not how much autonomy a tool has, but what kind of engine powers it. If you want the full breakdown, see [What Is Agentic AI? full guide]. The short version relevant here: generative AI is the part that produces an answer or piece of text on request, the same way a chatbot might. Agentic AI adds the memory and tool-use layer on top of that — the part that lets a system stay on task across multiple steps instead of resetting after each reply. That memory-and-tools layer is exactly what separates an assistant from an agent in the table above, and it’s covered in more technical detail in our [How do AI agents actually work?] post.
Which One Are You Actually Using?
A quick way to tell, next time you’re not sure what you’re dealing with: ask it to do something that takes more than one step, without giving it further instructions along the way. If it does the whole thing and reports back, you’re using an agent. If it does the first step and waits, that’s an assistant. If it just replies with information and does nothing, that’s a chatbot no matter what the product page calls it.