It is eleven at night. You write to the garage: "The oil warning light has come on, can I bring the car in first thing tomorrow?". The reply arrives in a second: "Hello, choose an option: 1) Opening hours 2) How to get here 3) Book an appointment". None of them quite fits. You write "appointment for early tomorrow" and it answers: "I did not understand your message. Return to the main menu". You write "3". It asks for your name. You give it. It throws the main menu at you again. You close the conversation and phone in the morning, when they open, as you have always done.
What answered you is sold as a "chatbot with artificial intelligence". It has none. It is a button tree: a list of fixed answers linked together, and as soon as you step off the script, it gets lost. An AI agent is something else. And the confusion between the two, fed by suppliers who call anything "AI", leads many businesses to pay for a menu of options believing they are buying something that understands their customers.
In this article we explain the real difference, with examples from the same business, so you can tell one from the other before you sign anything.
What is a classic chatbot?
A classic chatbot works with fixed rules. Someone sits down and draws a tree: if the customer presses "Opening hours", the bot gives the opening hours; if they press "Book", the bot asks for date, time and number of people in a closed order. Every branch is written by hand in advance.
It works well as long as the customer stays in the lane. The problem is that people do not write in lanes. They write the way they speak: they mix two questions in one sentence, give a detail before being asked, change their mind halfway, use a word that is not on the list. As soon as that happens, the button tree has nothing to go on. It replies "I did not understand you" or sends you back to the start. It is not broken: it can only do what it was drawn to do, and nobody can draw every way a customer asks for an appointment.
That does not make it useless. For a very simple menu (opening hours, address, a link) a button tree does the job and is cheap. The problem begins when it is sold as something that understands the customer, because it understands nothing: it recognises buttons.
What is an AI agent?
An AI agent starts from somewhere else. Instead of a list of buttons, it has a language model behind it, and that changes four things at the root.
- It understands natural language. It does not need the customer to choose an option. "Can you do tomorrow early?", "do you have a slot first thing?" and "I need to bring the car in before nine" are, for the agent, the same intent. It resolves all three the same way.
- It checks your system. An agent does not answer from memory: it looks. It checks the diary before offering a time, looks at the CRM to know whether the person writing is a regular customer, checks whether there is stock before promising anything. It answers with real data from your business, not with a generic reply.
- It carries out actions. It does not just inform. It books the appointment and writes it in the calendar, logs the order, updates the customer record, sends the confirmation. It does the work, not just describes it.
- It hands over to a human when needed. When the enquiry goes beyond what it can or should resolve (a serious complaint, an unusual case, anything involving money), it passes the conversation to a person with the context already collected, instead of crashing into an "I did not understand".
The underlying difference is this: a classic chatbot replies within a script; an AI agent understands, checks, acts and knows when to step aside. We go into more detail in what an AI agent is and why SMEs already use them.
The same enquiry, two outcomes
Back to the garage in the opening scene. A customer writes, at whatever hour: "The oil warning light has come on, can I bring the car in first thing tomorrow?".
The button tree does not find that sentence on its list. It offers the menu. The customer insists in normal language, the bot does not recognise it, sends them back to the start. At best, the customer gets to "Book an appointment" and the system writes a name down somewhere; at worst, they get tired and leave. Nobody has checked whether there is a slot first thing tomorrow, because the bot does not know how to look at the diary.
The AI agent does something else. It understands that the customer wants an appointment early tomorrow and that they mention an oil warning light. It looks at the garage's diary, sees there is a slot at 8:30 and offers it. The customer says yes. The agent creates the appointment, writes it in the calendar, collects the registration, leaves a note with the symptom the customer described so the mechanic arrives with the clue, and sends the confirmation. When the garage opens in the morning, the appointment is already in place and documented. If the customer had written "the car is smoking and smells of burning, I don't know if I can drive it", the agent would not try to diagnose: it would pass the alert to a person, because that is not solved over WhatsApp.
Change the sector and the pattern repeats. At a clinic, someone writes "I need to change Thursday's appointment, something has come up". The button tree has no button for "change Thursday's appointment"; at most it would have "manage appointment" and a rigid form. The agent finds that patient's Thursday appointment, offers the free slots the following week, reschedules and notifies. One sentence, resolved.
That is the leap: it is not that the agent replies faster, it is that it does the work the button tree only pretended to do.
"It's not a button tree with another name"
Here is the problem in the market. Since "AI" sells, some suppliers take an old-fashioned button tree, paint the words "artificial intelligence" over it and charge for it as if it understood customers. It understands nothing. It is still a menu of options in a different outfit.
At Xora Intelligence we build real agents, and we say it plainly: it is not a button tree with another name. An agent understands what is written to it, checks your real systems and carries out actions in them. That means integration work with your diary, your CRM or your management software, and that work is precisely what separates an agent from a disguised menu. If it does not touch your systems, it is not doing your work; it is showing buttons.
It is not a matter of fashion or of buying the most expensive thing. For some businesses a simple menu is all they need, and we will tell you so. But if they sell you "AI", it should be AI, and not a button tree with a new label.
How can I tell whether what I'm offered is an agent or a disguised tree?
You do not need to be technical. A few questions to the supplier show straight away which side of the line what they are selling falls on:
- "What happens if the customer does not use any of the options and writes in their own way?" If the answer is "they have to be guided with buttons", it is a tree. An agent understands the loose sentence.
- "Does it check my diary or my CRM in real time, or does it answer with fixed texts?" If it only gives prepared texts, it is not looking at your business; it is reading a script.
- "Does it carry out the action (create the appointment, log the order) or does it only alert me so I do it?" An agent does; a tree, at most, picks up a detail and forwards it to you.
- "What does it do when it cannot resolve something?" The good answer is "it passes it to a person with the context". The bad answer is "it returns to the main menu".
- "Which of my systems does it integrate with?" If the answer is "none, it runs separately", you are looking at an island that does not touch your business.
The same logic applies to phone service. A voice agent that answers the phone and understands what you say has nothing to do with the old "press 1, press 2" switchboard; what separates the agent from the button tree also separates a voice agent from a phone menu. And if your enquiries come in as text, the one we build as a chatbot is an agent, not a menu of options with the label changed.
What does the law say about telling people they are talking to an AI?
One important thing, whether it is an agent or a chatbot: you have to tell the customer. According to the European Commission's EU Artificial Intelligence Act (AI Act, Reglamento Europeo de Inteligencia Artificial), from August 2026 companies that use AI systems to converse with people are required to inform them that they are interacting with an artificial intelligence system, unless it is obvious. The Regulation entered into force on 1 August 2024 and is fully applicable from 2 August 2026.
In practice it is simple: the agent introduces itself as what it is at the start of the conversation and makes clear how to reach a person. It is not an obstacle, it is transparency, and it is how we build any conversational system. Hiding that there is an AI behind it does not only break the rule: it erodes the customer's trust, which is exactly the opposite of what you want.
So which do I need?
It depends on what it has to do, not on which sounds more modern.
If all you want is for a message to answer the opening hours, the address and little else, a simple menu does the job and there is no need to complicate things. If what you need is for someone, at any hour, to understand an enquiry written in free text, look at your diary, really book, log the order in your system and know when to pass you a conversation, a button tree cannot do that however much they call it "AI": you need an agent.
The honest way to decide is to look at what your customers ask you on WhatsApp or by phone for a whole week. If almost everything is four repeated questions, a menu is enough. If people ask for specific things, change appointments, mix requests and constantly go off script, a menu will let you down just when you need it most.
Frequently asked questions
Is an AI agent just a more expensive chatbot?
No, it is a different technology. A classic chatbot follows a script of buttons and gets lost as soon as the customer goes off it. An AI agent understands natural language, checks your systems and carries out real actions. That both are sometimes called "chatbot" does not mean they do the same thing.
How do I tell a real agent from a button tree with an "AI" label?
Ask what happens when the customer writes in their own way without using buttons, whether it checks your diary or your CRM in real time, and whether it carries out the action or only alerts you to do it yourself. If the answer is that the customer has to be guided with options and that it touches none of your systems, it is a button tree, whatever it is called.
Do I have to tell my customers they are talking to an AI?
Yes. Under the EU AI Act, from August 2026 you must inform people that they are interacting with an artificial intelligence system when it is not obvious. In practice, the agent introduces itself as what it is and makes clear how to reach a person. We build it that way by default.
Will an AI agent leave my team without work?
That is not its job. The agent absorbs the repetitive (the same questions, routine appointments) and passes to a person what needs judgement. Your team stops being tied to the phone or WhatsApp for what anyone could answer, and deals with what does need a human head.
Not sure whether your case calls for an agent or whether something simpler is enough? Tell us what you need and we will say whether we can do it, how we would approach it and what it would cost. If a simple menu works for you and nothing more is needed, we will tell you that too.
Xora IntelligenceAI consultancy and custom software in Los Barrios (Cádiz). We write about what we build for businesses in the province. If you have a question about your case, write to us.




