AI agent examples: 11 customer conversations and the 5 types of AI agents
Muhammad Hassan, FounderLinkedIn
Published Updated
The AI agent examples below are eleven customer conversations, each ending in something finished: a booking, a reservation or table request, a demo, a lead, an answered order question, an urgent ticket or a handover to a person. Each one shows the agent's replies, the outcome and what the business's team still confirms. After them come the five types of AI agents, with an everyday example of each.
We make HiQBot, a customer conversation platform whose AI agent answers customers on website chat, WhatsApp, Instagram, Messenger, Telegram and LINE. Every screen here is a real HiQBot screen with a sample business and a sample customer. The agent's lines follow what HiQBot's industry-tuned agent does for that kind of business. None is a transcript from a real customer.
What an AI agent is
An AI agent is software given a goal and the tools to reach it. It reads a request, works out what is missing, asks for it, then takes the step that finishes the job, such as writing a booking or opening a ticket. A chatbot that only answers questions stops before that last step.
Our guide to AI agents versus chatbots covers the difference in full. The first eleven examples come from customer service, where most of a business's customers will meet an AI agent first. Each one names the channel, what the agent did, the outcome, and the part a person still owns.
Examples of AI agents in customer service
1. A salon agent books a new colour client, patch test first
A hair salon on WhatsApp at lunchtime. Hannah asks the balayage price and whether she can come in on Saturday.
- What it did
- Gave the starting price and what changes it, asked whether she was new, then asked for the patch test at least 48 hours before the colour, as the salon asks of new colour clients, with the stylist her friend recommended on the booking.
- Outcome
- Two bookings: the patch test on Tuesday and the balayage on Saturday.
- The team
- Confirms both bookings and handles the $30 colour deposit. The agent only mentions it; it never takes payment.
2. A clinic agent books a new patient on a Sunday night
A family clinic on WhatsApp at 8:40 on a Sunday evening. Sarah wants a first appointment for her 7-year-old son.
- What it did
- Asked whether the family was registered and which doctor they wanted, noted the insurer’s name without asking for a policy number, said the front desk checks cover on arrival, and asked for the free slot with Dr Clarke. It never asked about symptoms.
- Outcome
- A new patient appointment on Tuesday at 3:40 PM.
- The team
- The front desk confirms the appointment from the bookings list and checks the insurance when the family arrives.
3. A real estate AI agent turns a 9:40 PM listing question into a showing
A real estate team’s website chat late in the evening. Lauren asks whether 212 Willow Bend Drive is still available and what the HOA is.
- What it did
- Answered the price, bedrooms and HOA from the listing, said the listing agent would confirm it was still available, asked whether she was buying to live in and had a lender, then offered the open house or a private showing.
- Outcome
- A private showing on Saturday at 11 AM with a buyer pre-approved up to $500,000.
- The team
- Megan confirms the home is still available and confirms the showing.
4. A hotel agent files a reservation request and a table request
A seaside hotel on WhatsApp at 10 PM. Isabel wants a family room with a sea view for three nights over her anniversary.
- What it did
- Said what the room sleeps, the nightly rate with breakfast and the parking charge, filed a reservation request with her late arrival noted, then offered a table for the first evening.
- Outcome
- A reservation request for 20 to 23 November and a table request for 7:30 PM, marked as an anniversary.
- The team
- The front desk confirms the stay by email; the restaurant confirms the table. The agent never says a room is booked.
5. A restaurant agent takes a birthday table with a coeliac guest
A restaurant on WhatsApp at 10:41 PM. Lucy needs a table for six on Friday for her mum’s birthday; her mum is coeliac.
- What it did
- Offered the two times it could request, said what the gluten-free menu covers, and wrote the allergy onto the request in Lucy’s own words.
- Outcome
- A table request for six on Friday at 7:30 PM with “My mum is coeliac” for the kitchen.
- The team
- Confirms the table in the same chat. The kitchen reads the allergy note; the agent never promises a dish is safe.
6. An auto repair agent turns a price question into a drop-off
An independent repair shop’s website chat at 9:12 PM. Ryan asks what front brakes cost on his 2017 Honda Accord, which squeals when he stops.
- What it did
- Gave the starting price from the shop’s price list, said the final price depends on what the technician finds, asked the mileage and offered two drop-off times.
- Outcome
- A brake inspection request for Tuesday at 8 AM, with the squeal and the mileage in the notes.
- The team
- Jenna at the front desk confirms the drop-off.
7. A personal trainer’s agent takes a partner training enquiry
A personal trainer’s website chat at 10:41 PM. Daniel asks whether the trainer does partner sessions.
- What it did
- Said partner sessions are priced per pair, then asked what the pair wanted to work on and their monthly budget. Daniel replied with all of it, plus the start date and his email.
- Outcome
- A lead: partner training for two beginners, Saturdays from January, about $400 a month.
- The team
- Sam, the trainer, emails his options.
8. A software company’s agent answers a pre-sales question and logs a demo
A software company’s website chat at 10:41 AM. Megan wants to know whether it syncs with HubSpot before she books a demo.
- What it did
- Answered the HubSpot question from the company’s own documentation, then, when Megan described 25 technicians scheduled from a spreadsheet, offered three demo times and asked for her work email and company.
- Outcome
- A demo request for Tuesday at 1:30 PM Eastern for Kestrel Home Services.
- The team
- Sales confirms the slot and runs the demo; the team size and the spreadsheet problem are in the chat.
9. An online store’s agent answers “where is my order?”
An online clothing store on WhatsApp at 6:21 PM. James asks where order 519 is.
- What it did
- Asked for the email or phone number used at checkout before saying anything about the order, then read back the status from the connected store: cash on delivery, not shipped yet, one hoodie, $77.99 due to the courier.
- Outcome
- The question answered in under a minute, with no ticket and nobody on the team involved.
- The team
- Nothing, unless James asks to cancel, exchange or change the address; those come to the team to approve.
10. A property manager’s agent opens an urgent repair ticket at midnight
A property management company on WhatsApp at 11:48 PM. Natalie reports water coming through her bathroom ceiling.
- What it did
- Asked whether the water was near lights or sockets. When it was, it told her to call the out-of-hours line now, and wrote the repair up as an urgent ticket with the flat, the address and the leak, and gave her the reference.
- Outcome
- An urgent ticket, TKT-8a41c2e7, with the tenant on the phone to the out-of-hours line.
- The team
- The out-of-hours line deals with tonight. The ticket holds the address and what she has already done, ready for the maintenance team.
11. A repair shop’s agent hands a question it cannot answer to a person
The same repair shop on WhatsApp at 2:11 PM. Kevin asks whether his pickup truck, dropped off that morning for an oil leak, is ready.
- What it did
- Said it couldn’t see how the work was going and passed the chat to the front desk instead of guessing.
- Outcome
- A handover. Jenna replied three minutes later in the same chat, with the whole conversation in front of her.
- The team
- Answers in the HiQBot inbox with the whole chat above her reply.
What each agent finished, and what the team does
Every booking the agent takes lands in the dashboard against the customer, with Confirm, Decline and Reschedule beside it, and on your Google Calendar if you connect one. The agent never takes payment. The clinic appointment from example 2 sits in the list below, waiting for the front desk.
| Example and outcome | The team |
|---|---|
| Salon, WhatsAppTwo bookings | Confirms both bookings |
| Clinic, WhatsAppAppointment booked | Confirms, checks cover on arrival |
| Realtor, website chatShowing booked | Agent confirms availability and the showing |
| Hotel, WhatsAppReservation and table requests | Front desk and restaurant confirm |
| Restaurant, WhatsAppTable request | Confirms the table in the chat |
| Auto repair, website chatDrop-off request | Front desk confirms the drop-off |
| Personal trainer, website chatLead captured | Trainer emails his options |
| Software company, website chatDemo request | Sales confirms and runs the demo |
| Online store, WhatsAppOrder question answered | Nothing unless a change is asked |
| Property manager, WhatsAppUrgent ticket | Out-of-hours line, then maintenance |
| Auto repair, WhatsApp handoverHandover to a person | Replies in the same chat |
Types of AI agents, with an example of each
Russell and Norvig's textbook Artificial Intelligence: A Modern Approach describes four basic kinds of agent plus learning agents, and IBM's guide to AI agent types lists the same five, from the rule-following to the self-improving. Sources read on . A real product usually mixes several: the agents above follow rules, remember the conversation and work toward a goal at the same time.
1. Simple reflex agents
Follows fixed if-this-then-that rules on what it sees right now, with no memory. Everyday example: a thermostat that switches the heating on below a set temperature.
In customer service: a guided flow, where the customer taps “Opening hours” and gets the hours. HiQBot’s guided flows work this way: buttons and steps, with the AI only where you switch it on.
2. Model-based reflex agents
Keeps a picture of what has happened so far and uses it with the rules. Everyday example: a robot vacuum that remembers where the furniture is.
In customer service: the salon agent kept that Hannah was a new client recommended to Jess, and put Jess on the booking.
3. Goal-based agents
Plans the steps that reach a goal. Everyday example: a route planner finding a way across town.
In customer service: every example above. The goal is a booking, a request, a demo or a ticket, and the agent asks only what that goal still needs.
4. Utility-based agents
Scores the options and picks the one with the best trade-off. Everyday example: a driverless car weighing speed, comfort and safety.
In customer service: rare. Suggesting the product that best fits a budget comes closest.
5. Learning agents
Gets better from feedback and experience. Everyday example: a streaming service whose recommendations improve as you watch.
In customer service: HiQBot’s agent does not retrain itself on your chats. It lists the questions it could not answer; you answer each once and it uses that answer from then on.
Two more terms come up often. A multi-agent system splits a job between several agents, for example one that sorts incoming messages and another that books. A hierarchical agent is the same idea with a manager agent handing tasks down to the others.
AI agent examples outside customer service
The same idea, a goal plus the tools to reach it, runs well beyond customer chat.
| Job and what the agent does | Example |
|---|---|
| Writing codeA coding agent takes a task and returns the change for review. | GitHub Copilot: you assign it an issue, it works on a branch and opens a pull request for a developer to review (GitHub Docs). |
| DrivingA driverless car reads its sensors and chooses how to move. | Waymo's driverless ride-hailing cars. |
| Research and the webAn assistant with a browser looks things up, compares them and fills in forms. | AI assistants that can open web pages and act in them, usually asking before they submit anything. |
| Home and devicesA device acts on readings from a sensor. | Smart thermostats and robot vacuums. |
| RecommendationsA system picks what to show next from what you did before. | Shopping and streaming recommendations. |
What the good examples have in common
- They answer the question first, from the business's own prices, listings or policies, then ask for what the next step needs.
- They ask one or two things at a time and skip anything the customer already said.
- They write what they learn onto the lead, so the team reads the goal, budget or allergy without scrolling the chat.
- They say plainly who confirms: the front desk, the restaurant, the realtor showing the house.
- They hand the chat to a person when the answer is not theirs to give, like the repair shop in example 11.
- They reply in the customer's language, and at 8:40 on a Sunday evening as readily as at noon.
For more conversations, without the screens, our chatbot examples cover ten industries plus FAQ bots, a chatbot script and a feedback question.
Where AI agents go wrong
- Inventing a price. The repair shop agent gave the starting price from the shop's list and said the inspection settles the rest.
- Promising what only a person can promise. The hotel and restaurant agents file requests and say the team confirms; they never say a room or a table is held.
- Giving advice it should not give. The clinic agent takes the appointment request and keeps symptoms out of the chat. Anyone describing an emergency is told to call the emergency services first.
- Handling money. The agent sends a checkout link or files an order for the team to confirm. It does not take a payment in the chat.
- Holding on too long. The property manager's agent sent the tenant to the out-of-hours line within two messages, because water near a light can't wait for a ticket.
How to set up your first AI agent
- Pick one job, such as bookings, table requests or order questions, and get that right first.
- Give the agent your prices, services, opening hours and policies as files or website pages.
- Connect the channels your customers already write on, starting with the busiest.
- Decide who takes over when the agent hands a chat to the team, and who confirms bookings.
- Read the first week of conversations and answer the questions the agent could not.
Our guide to AI agents for small businesses goes through choosing one and what to watch for, and the AI customer service agent page shows what HiQBot's agent does.
Common questions
What is an example of an AI agent?
A WhatsApp agent that answers a salon client's price question and takes the booking request for her patch test and appointment is an AI agent. So is a coding assistant that takes an issue and opens a pull request, or a driverless car.
What are the 5 types of AI agents?
Simple reflex, model-based reflex, goal-based, utility-based and learning agents. The list builds on Russell and Norvig's AI textbook, and most real products combine several of them.
Is ChatGPT an AI agent?
Not when it only answers in a chat; that makes it an AI assistant. It acts as an agent when it is given tools to take steps for you, such as browsing a website or running code.
What is the difference between an AI agent and a chatbot?
An AI agent takes the step after the answer. In the examples above it answered the question, then wrote the booking, the request or the ticket.
What are examples of AI agents in customer service?
Booking an appointment on WhatsApp, filing a hotel reservation request, answering “where is my order?”, logging a demo request, opening an urgent repair ticket and handing a chat to a person. All eleven are shown above.
Can an AI agent answer phone calls?
Some can: voice AI agents answer phone calls. HiQBot's AI agent works in writing and never answers or makes a call. Customers can still call your WhatsApp Business number, and the call rings your team in the HiQBot inbox, beside the chat.
Which channels can a customer service AI agent work on?
HiQBot's agent works on website chat, WhatsApp, Instagram (including DMs started from a post comment), Messenger, Telegram and LINE, with every conversation in one inbox.
What is an example of a multi-agent system?
A support setup where one agent sorts incoming messages and passes each to another agent for billing, bookings or technical questions. Each agent does one job and hands the rest on.