What is an AI agent?

Software that understands a goal in plain language, works out the steps and carries them out with the tools it has been given.

Short answer

An AI agent is software that takes a goal written in plain language, decides which steps are needed and carries them out with the tools it has been given.

A chatbot replies to a message. An AI agent finishes a task: it reads the request, looks up what it needs, picks an action such as searching a catalogue or booking a slot in a calendar, checks the result and carries on until the job is done or a person has to take over. The language model does the understanding, the tools do the work, and the rules set by the company decide what it may and may not do. In customer service that means an agent that answers a question from the company's own pages, looks up the order, books the appointment, writes down the lead and passes a complaint to a colleague with the whole conversation attached. SimplyBoost is an AI agent of that kind, for website chat, WhatsApp and Instagram.

  • Explained without jargon
  • Questions people really ask, answered one by one
  • Updated November 2026

By , AI EngineerUpdated

How does an AI agent work?

Every AI agent runs the same short loop, whether it plans a trip, writes code or answers a customer. The difference between agents is which tools they are allowed to use and where they must stop.

  1. 1. It reads the request

    A language model works out what the person wants, even when the wording is loose or in another language.

  2. 2. It looks things up

    It searches the sources it was given: pages, documents, a product catalogue, an order system.

  3. 3. It decides the next step

    Answer now, ask a question back, or use a tool. This choice is what makes it an agent.

  4. 4. It acts

    It books the slot, fetches the order, saves the lead or writes the reply.

  5. 5. It checks the result

    If the goal is met it stops. If not, it goes round again.

When it should not decide alone

A complaint, a refund or a question outside its sources goes to a person, with the conversation attached.

The loop of an AI agent. In SimplyBoost the sources in step 2 are your own website, documents and FAQs, and nothing else.

What is the difference between a chatbot and an AI agent?

A chatbot talks. An AI agent talks and then does something about it. The two are often mixed up because they share a chat window.

Scripted chatbotAI agent
How it understandsMatches keywords or buttons to a fixed scriptReads the whole message in plain language, in the customer's language
Where answers come fromReplies somebody typed in advanceThe company's own pages and documents, read at the moment of the question
What it can doShow the next step of a flowUse tools: look up an order, book an appointment, save and rate a lead, open a ticket
A question nobody planned forEnds in "I did not understand that"Answers if the sources cover it, and says so and hands over if they do not
Setting it upBuild and maintain a flow for every caseAdd your content and write the rules in plain words
Keeping it rightEdit the flowEdit the page or document, and the answer changes with it

Can a chatbot be an AI agent? Yes: when the thing behind the chat window can understand free text and take actions, it is an agent that happens to live in a chat. The AI chatbot page explains the chat side, and the AI support agent page shows the agent at work.

What does an AI agent do exactly?

It depends on the tools it has been given. These are the jobs a customer service agent does in a normal week, each one a task it finishes and not only a reply it writes.

Answers from your own content

It reads your website, PDFs and FAQs and answers from them. If the answer is not there it says so, and the question is saved so you can add the missing page.

Looks up an order

In a Shopify store it checks the order after an email code, shows the status and the tracking link, and can start a return. See the Shopify AI chatbot.

Books an appointment

In website chat it checks Google Calendar, offers the free slots and puts the booking in the calendar, without a form or a phone call.

Captures and rates a lead

After it has helped, it asks for a name and an email or phone number, scores the lead from 0 to 100 and marks it Hot, Warm or Cold. See lead capture.

Hands over to a person

A complaint, a refund or a request for a human goes to your team by email or Slack, or as a ticket in Zendesk, Freshdesk or Gorgias, with the conversation attached.

A person in a dark room late at night, their face lit by the phone they are typing a question on

23:10, website chat

Can I still move my appointment on Thursday to next week?

Yes. Tuesday 10:30 and Wednesday 14:00 are free. Which one shall I book for you?

What does an AI agent look like to a customer?

Like a good colleague who happens to be awake. The customer types a question the way they would say it, gets an answer that is right for this company, and the thing they wanted is done before they put the phone down.

  • No menu to click through and no "please rephrase".
  • The answer comes from the company's own pages, with the source shown underneath.
  • When a person is needed, the conversation moves to the team and nothing has to be repeated.

What are the 5 types of AI agents?

Computer science textbooks sort agents into five kinds, from the simplest to the most capable. Today's agents built on language models are a mix of the last three.

Simple reflex agents
React to what they see right now with a fixed rule. A thermostat that switches the heating on below 19 degrees is the classic example.
Model-based reflex agents
Keep a small picture of the world, so they can act on things they cannot see at this moment. A robot vacuum that remembers which rooms it has cleaned.
Goal-based agents
Are given a goal and plan the steps to reach it. A route planner that finds a way from A to B belongs here, and so does an agent that books an appointment.
Utility-based agents
Weigh several possible outcomes and pick the best one, not only one that works. A route planner that balances time, tolls and traffic.
Learning agents
Get better from feedback. A customer service agent that shows which questions it could not answer, so the company can add the missing content, has this loop built in.

Is ChatGPT an AI agent?

ChatGPT is a general AI assistant. It answers from what its model learned and from what it finds on the web, and in its agent mode it can also carry out tasks such as browsing a site or filling in a form. So parts of ChatGPT are an AI agent, and the chat most people know is an assistant.

The more useful question for a business is whose agent it is. ChatGPT works for the person typing. It does not know your prices, your opening hours or your return rules unless they happen to be on the web, and it cannot look up one of your orders. An AI agent for customer service is set up for one company: it answers only from the content that company added, on the company's own channels, under rules the company wrote, and it hands over to the company's team.

The two work well side by side. People ask ChatGPT which product or supplier to choose, then come to your website or WhatsApp with the detailed questions, and your own agent answers those. The page on ChatGPT for customer service goes deeper into the difference.

What should you look for in an AI agent for your business?

It stays inside your content

  • Answers only from the sources you added
  • Shows the source under each answer
  • Says it does not know, and saves the question, when the answer is missing

It can actually do things

  • Looks up orders and recommends products in a webshop
  • Books appointments in a calendar
  • Captures and rates leads, and exports them

It knows when to stop

  • Hands over on complaints, refunds and requests for a person
  • Follows rules you write in plain words
  • Leaves your team in charge of every judgement call

How do you make an AI agent for your business?

You do not need to write code or train a model. With SimplyBoost it takes about five minutes, and the full guide shows every screen.

  1. 1

    Create an account

    Sign in with a work email or Google. Your first agent is created for you.

  2. 2

    Add your content

    Type your website address and the agent reads the pages. Add PDFs or text for anything that is not on the site.

  3. 3

    Give it a role and rules

    A name, a tone, and the things it must never do, written in plain words.

  4. 4

    Test it with a real question

    Ask something from last week's inbox in the preview and check the answer and its source.

  5. 5

    Put it where customers are

    On your website with the WordPress plugin, the Wix or Shopify app or one script line, then on WhatsApp and Instagram.

Questions people ask about AI agents

An AI agent is a program you give a goal to, in ordinary language, and that then works out the steps and carries them out by itself. It can look things up, use tools such as a calendar or an order system, and check whether the job is done. A chatbot only replies, an AI agent finishes the task.

It reads a request, finds the information it needs, decides on the next step, acts with a tool and checks the result, in a loop, until the goal is reached. In customer service that means answering a question from the company's own content, looking up an order, booking an appointment, saving a lead or passing the conversation to a person.

Partly. The ChatGPT most people use is a general AI assistant that answers from its training and from the web. Its agent mode can carry out tasks, which makes that part an AI agent. It works for the person typing, though, and it does not know a company's own prices, orders or rules. A customer service agent such as SimplyBoost is set up for one company and answers only from that company's content.

A scripted chatbot follows a fixed flow of buttons and keywords and stops when a question was not planned for. An AI agent understands free text, finds the answer in the company's own content and takes actions such as booking an appointment or looking up an order. A chatbot talks, an AI agent talks and then does the work.

An AI assistant helps one person with what they ask for, step by step, and waits for the next instruction. An AI agent is given a goal and works through the steps on its own, inside the limits it was given. The same language model can power both. The difference is how much the software may decide and do without asking.

The textbook classification has five: simple reflex agents that follow fixed rules, model-based reflex agents that keep a picture of the world, goal-based agents that plan towards a goal, utility-based agents that pick the best of several outcomes, and learning agents that improve from feedback. Agents built on language models combine the last three.

It takes the repeated work that has a known answer. In customer service it answers questions at any hour from your own content, looks up orders, books appointments, captures and rates leads, and hands the difficult conversations to your team with the context attached. Your team keeps the judgement calls and loses the repetition.

It depends on the pricing model. Some tools charge per seat, some per resolved conversation, and some a flat monthly price. SimplyBoost is flat: €39 a month for 500 AI replies, €89 for 2,000 and €169 for 4,000, with website chat, WhatsApp and Instagram in every plan and custom plans above that. See the pricing page.

General assistants often have a free tier. An AI agent that works for a business on its own channels is normally a paid service, because every answer costs computing time. SimplyBoost starts with a 7-day free trial that includes 100 AI replies, so you can test it on your own content before you pay.

Pick a platform that does not need code, add your website and documents as sources, write the rules in plain words, test it with real customer questions and put it on your website and messaging channels. With SimplyBoost that takes about five minutes. The step-by-step guide shows every screen.

A language model on its own can. A well-built agent is kept to named sources, a method called retrieval-augmented generation. SimplyBoost answers only from the content you added, shows the source under the answer, and when the answer is missing it says so and saves the question under Missing answers.

They replace the repetition, not the people. An agent handles the questions that have a written answer, at any hour and in any language the customer writes in. Complaints, exceptions and anything that needs judgement go to a person, who now has the time for them. Most teams use the agent as the first line and keep people as the second.

The best one for your company answers only from your own content, works on the channels your customers already use, hands over to a person with the whole conversation, and has a price you can predict. SimplyBoost does all four, with a flat monthly price and no per-seat or per-resolution fees. The comparison pages set it next to other tools.

This page in other languages: Nederlands, Deutsch, Français, Español, Italiano, Polski, Suomi

Try SimplyBoost with your own content

Start a 7-day free trial with 100 AI replies. Add your website, install the widget and see how the agent answers your customers. Larger team? Custom plans above 4,000 AI replies come with a dedicated account manager and white-label.

See pricing
  • 7 days, 100 AI replies
  • From €39 a month
  • Hosted in Europe