Your customer types "where's my order?" into the chat on your website. One tool replies with a link to your shipping policy. Another looks up their actual order, sees it left the warehouse yesterday, gives them the tracking number and tells them it should arrive Thursday. Both get called "AI". Only one of them is an agent, and your customer can tell the difference straight away.
A chatbot answers questions. It matches what someone typed to an answer it already has, usually from a script, a decision tree or a set of FAQs. The newer ones use AI to sound more natural, but they still mostly talk.
An AI agent works towards an outcome. It reads the question, looks up the records it needs, decides what to do, uses the tools it's been given (update a record, create a ticket, book a time, draft an email) and hands over to a person when it should. It does things, not just says things.
If you want the longer explanation of what makes something an agent, start with our plain-English guide, What Is an AI Agent?
When a vendor shows you their "AI assistant", ask these. The answers sort chatbots from agents quickly.
A chatbot usually knows your business in general: your hours, your policies, your services. An agent knows this customer: their open quote, their last invoice, the ticket they raised last week. That only works if the agent is connected to a CRM where all of that lives on one record and is joined together properly.
"I've rebooked you for Tuesday at 2pm and sent the confirmation" is an agent. "You can rebook using this link" is a chatbot. Both have their place, but only one removes work from your team.
A weak chatbot guesses, loops or says "I didn't understand that." A well-built agent knows its limits, creates a ticket with the conversation attached, routes it to the right person and tells the customer what happens next. The hand-off is designed, not an afterthought.
Because an agent writes to the CRM, the next person (or the next agent) can see what was said and done. A chatbot conversation often lives in its own silo and disappears.
Customers don't care what the technology is called. They care whether their problem got solved. A chatbot that answers with a generic link is fine for "what are your opening hours?" It's frustrating for "I was charged twice." That second customer now has to repeat themselves to a person, and they're already annoyed.
An agent can carry that customer further. It can see the two charges, confirm the duplicate, create a ticket for your accounts person with the details already filled in, and tell the customer it's been flagged for a refund today. A person still approves the refund, because money stays with a person. But the customer has been heard, and your team has a ready-to-action ticket instead of a vague complaint.
That's the shift. Chatbots deflect questions. Agents move work forward. We covered how this plays out in service teams in AI in customer service, and the principle hasn't changed: support is only as good as the data behind it.
Here's the uncomfortable bit. Lots of products sold as agents can't do much more than a chatbot, because they can't reach good data. They're plugged into a website and a PDF of FAQs. They sound clever, but they can't look anything up.
The model is rarely the problem. The data layer is. We wrote a whole post on this, Why Most AI Agents Fail: It's the Data, Not the Model. The short version: an agent needs a structured place to look. That's why we build agents on HubSpot. Contacts, companies, deals, tickets, orders, invoices and custom objects all sit in one CRM, and the associations between them tell the agent which records belong together. When a customer asks about their order, the agent follows the path from contact to order to shipment and gets the real answer.
Not every business needs an agent on day one. A chatbot (or HubSpot's chatflows) is often enough when:
An agent starts to earn its keep when:
Our after-hours example for trades, The After-Hours Receptionist, shows that line clearly. A chatbot takes a message. An agent books the job.
Customers will forgive an agent for saying "let me get someone to help with that". They won't forgive one that confidently gets it wrong. So we build in the guardrails from the start:
HubSpot's own Customer Agent covers a lot of this out of the box, and it's usually where we start. We only build custom where it stops. The HubSpot agents page explains how we combine the two.
Next time you see a demo, ask it about a real customer's real order. If it can find the answer and do something about it, you're looking at an agent. If it sends you a link, it's a chatbot. Both can be useful. Just make sure you're paying for the one you need. If you've got a few spare minutes, our list of AI automations every HubSpot user should set up covers the quick wins that sit either side of that line.
Want to know whether your customers need a chatbot or an agent? Book an AI Readiness Audit and we'll look at the questions your customers actually ask, check whether your HubSpot data can support an agent, and recommend the right starting point. Already know what you need? Go to Build my solution.