What is the difference between an AI agent and a chatbot?
A chatbot answers questions. An AI agent answers questions and then does something about them: it checks your calendar, books the appointment, updates the customer's record, sends the confirmation text and follows up if they go quiet. The difference is whether the conversation ends with a finished task or a link.
Most local business owners have met the old kind of chatbot. It pops up on a website, offers four buttons, and when your question doesn't fit a button it tells you to call during business hours. Newer chatbots use AI to write better replies, but many still stop at the reply. An agent is connected to the systems that run your business, so it can act inside them.
How does a regular chatbot work?
A regular chatbot follows a script or searches a set of answers. It matches what the visitor typed to the closest answer and sends it back. It can't see your schedule, your customer list or your prices unless someone typed them into its answers, and it can't change anything.
There are two common types:
- Rule-based bots. A decision tree with buttons. Cheap and predictable, and useless the moment someone asks something off-script.
- AI FAQ bots. A language model reads your FAQ page and writes an answer in plain English. Better conversation, same dead end: "Here's our booking link."
Both are fine for answering "what are your hours" at midnight. Neither books the job.
What can an AI agent do that a chatbot can't?
An AI agent can take actions across your tools in the middle of a conversation: book and reschedule appointments, qualify a lead with your questions, update your CRM, send texts and emails, send a payment link, and hand off to a person with the full history attached.
| Task | Chatbot | AI agent |
|---|---|---|
| Answer common questions | Yes, from a script or FAQ | Yes, from your FAQs and business details |
| Book an appointment | Sends a booking link | Checks open times and books it in the conversation |
| Qualify a lead | Rarely | Asks your qualifying questions and tags the lead |
| Update your CRM | No | Creates or updates the contact and pipeline stage |
| Follow up later | No, the chat just ends | Texts or calls back on a schedule you set |
| Answer phone calls | No | Yes, with a voice agent |
| Hand off to a human | "Please call us" | Passes the conversation and notes to your team |
A lot of local leads still call. A voice agent picks up, has the same conversation a text agent would, and books the call into your calendar. Our breakdown of what AI voice agents cost and whether they are worth it has the numbers.
Is "AI agent" just a new name for a chatbot?
Sometimes, yes. Gartner calls it "agent washing": vendors relabeling chatbots and automation tools as agents without adding real agent abilities. Gartner estimates only about 130 of the thousands of vendors selling agentic AI are the real thing.
That estimate comes from Gartner's June 2025 prediction that more than 40% of agentic AI projects will be canceled by the end of 2027 because of rising costs, unclear business value or weak risk controls. Gartner's advice was to use agents only where they clearly pay off.
For a local business, "clearly pays off" is easy to measure. Count the leads that came in after hours or while your team was on a job, and count how many got booked. If the tool you're paying for can't move that second number, it doesn't matter what the vendor calls it.
Which one does a local business actually need?
If people only ask you simple questions and book on their own, a chatbot is enough. If leads come from ads, calls or texts and need someone to answer, qualify and book them fast, you need an agent, because that's the work a chatbot hands back to you.
A quick way to decide:
- A chatbot fits if you sell online, most questions are about shipping or hours, and customers check out without talking to anyone.
- An agent fits if you run ads, get phone calls, quote jobs, book appointments or consultations, or lose leads because nobody answered in time.
- Both is common. The same agent can answer the FAQ questions and also book the job, so most service businesses skip the plain chatbot entirely.
Speed is the reason agents win for service businesses. The Harvard Business Review lead response study found that companies that tried to contact leads within an hour were nearly seven times as likely to qualify them as companies that waited longer. A chatbot that sends a link doesn't contact anyone. Run your own numbers in our missed call calculator to see what slow answers cost you each month.
What should an AI agent be allowed to do on its own?
Let an agent handle the routine, repeatable work on its own: answering known questions, booking, rescheduling, reminders and follow-ups. Send anything with money at stake, an upset customer or a question it hasn't seen to a person, with the conversation attached.
Gartner predicts that by 2029 agentic AI will resolve 80% of common customer service issues without a human, and that this will cut operating costs by 30%. That covers common issues. Upset customers and unusual requests still need you.
Small details break agents too. In our own calendar data, a wrong time zone on one booking caused a no-show. Over 45 days, the calls our AI agents booked went 10 showed and 14 no-shows, and most no-shows were booked 2 to 5 days out, while most people who showed were booked within 24 hours. So we set agents to offer the soonest open time first and to confirm the time zone out loud.
How do you test whether a vendor's AI agent is real?
Give it a real job and watch whether it finishes. A real agent books the appointment, updates the record and confirms with the customer without a person stepping in. A relabeled chatbot sends a link or says someone will reach out.
- Text or call it as a customer at 9pm and ask to book for tomorrow morning.
- Ask a question that isn't on your FAQ and see whether it makes something up or hands off.
- Change your mind halfway and ask to reschedule.
- Check your calendar and CRM afterward. The booking, the contact and the notes should all be there.
- Wait a day without replying and see whether it follows up on its own.
- Ask how it hands a conversation to a person, and test it.
If it passes all six, it's an agent. If it fails the first one, it's a chatbot with a new label.
KenjiAI gives you both kinds of agent in one place: Conversation AI for website chat and text, and Voice AI for inbound and outbound calls. Both qualify the lead, book into your calendar, update the CRM and hand off to your team in a shared inbox. We set them up for you and test them against the six steps above before they talk to a real lead. See the voice agents page or book a call.
Where should you start?
Pull up last month's leads and mark every one that came in after hours, went to voicemail or waited more than an hour for a reply. That list is the work an agent would have done. Run the six-step test on whatever tool you use now, and if it fails, replace it starting with the channel where most of those leads came from.