← Back to the blog Article
AI agents · 9 min read

An AI agent is not a chatbot.
Here is the difference — and why it matters

Five years ago, chatbots drove you mad because they understood nothing. An AI agent is a different animal: it does not pick from pre-written replies, it reads your data — and it can also get things done. Below is what that means in practice, what it costs, when it pays for itself and when I would talk you out of it.

On Friday at 9.40 pm someone wrote to your Instagram inbox: “Hi, do you have anything free on Saturday?” You saw the message on Monday morning. By then they had already found someone else — not because you are worse, but because someone was faster.

This does not happen once. It happens every week — you just never see it, because a lost enquiry shows up in no spreadsheet. You only see the ones that stayed.

An AI agent is not a chatbot. The difference is bigger than it sounds

Over the past decade the word “chatbot” has earned itself a bad name, and deservedly so. The first chatbots were trees of pre-written replies: if the customer did not ask exactly what someone had predicted, they got a “sorry, I do not understand”.

An AI agent is something else. It does not choose from pre-written replies — it reads your data and builds the answer from it. More importantly, it can also take action: book an appointment, send a quote, log an enquiry, draft an email for you.

What it isWhat it can doWhen it breaks
Button chatbotWalks people down a pre-drawn path. Three buttons, five answers.The moment a customer asks something of their own.
AI assistantAnswers questions from your data — prices, schedule, process, FAQs.When something needs doing, not just saying.
AI agentAnswers and acts: books the appointment, prepares the quote, logs the enquiry, notifies you.When it hits a case nobody predicted — then it hands over to a human.

Most small businesses actually need the middle or the bottom row. The top one is what put people off.

What an agent actually does in a small business

No sweeping promises. These are the jobs an agent genuinely takes over:

Hair or beauty salon

Answers a question about prices and free slots at 9.30 pm, offers two times, books the appointment and sends a confirmation. In the morning you see a booked client, not fifteen unanswered messages.

Tradesperson — plumber, joiner, decorator

An enquiry arrives while you are on site: the agent asks for measurements, material and deadline, writes everything up in one format and drafts a quote for you by the evening. You correct a figure and send it. Half an hour instead of two.

Accountancy practice

Clients email the same questions about deadlines, documents and procedures. An agent trained on your internal notes answers the recurring ones and hands the hard ones to you with a ready-made summary.

Estate agent

An interested buyer asks at 11 pm about the floor area, the price per square metre and the energy rating. The agent answers from the property data and offers a viewing. You get a contact and their requirements, not a missed call.

Why people actually say yes — and it is not the technology

In these conversations, almost nobody ever says “I'm interested in artificial intelligence”. What happens is something else — an event. Usually one of four:

  1. The season. In summer or before the holidays enquiries double while the team stays the same. Someone has to answer at night.
  2. Someone leaves. The person who held the phone and the inbox moves on. Suddenly it is obvious how much work that was.
  3. A lost job that hurt. A customer went to a competitor because nobody replied for two days. This is the strongest trigger.
  4. A hiring decision. The owner compares a monthly salary with a one-off system build and runs the numbers for the first time.

If none of these sounds like you, you probably do not need an agent yet. I would rather tell you that now than in three months.

What an agent cannot do — and when not to buy one

Honest about the limits An agent does not replace a salesperson on big deals, where the decision is emotional and slow. It does not understand context that is not in the data — if your price list is written down nowhere, it cannot quote it. It does not fix a broken process; if things are chaos now, you get faster chaos. And it must not decide on its own where a mistake is expensive — there it always asks you.

Three cases where I would advise against it: fewer than ten enquiries a month, every job completely different from the last, or no prices or procedures written down anywhere. In the last case the first step is writing the process down, not buying a system.

What it costs and when it pays for itself

With me, an agent build starts at 990 € and takes one to two weeks. Add the AI model usage, which at lower volumes comes to a few tens of euros a month and is paid directly to the provider, not to me.

The maths is simple and you can do it without me. If you lose one customer a week because you could not reply in time, and an average job is worth 60 €, that is 3,120 € a year. Against a 990 € build, it pays for itself in four months — from rescued enquiries alone, before counting the hours saved.

Put your own numbers into the missed calls calculator. If it shows you less than a thousand euros a year, do not buy an agent.

How the build works

  1. A call. I tell you what is possible in your case and what is not. If it is not worth it, I say so.
  2. Gathering the material. Price list, FAQs, schedule, tone, boundaries — everything the agent needs to know. This is the only part that needs your time, and it takes an hour.
  3. Build and training. The agent learns your data; I embed it on your site or in WhatsApp.
  4. Testing together. You ask it the things your customers ask. Whatever it gets wrong, we fix.
  5. Handover and the first month. You get the access, the code and the instructions. For the first month we review the conversations together and fine-tune it.

The system is yours. Access, code and documentation are handed over at the end. A monthly retainer exists because some people want one, not because the system needs it to run.

You can try an agent right now

On my homepage, in the bottom right corner, there is an AI assistant that answers questions about my services. Ask it something hard — including what it cannot do. It is the fastest way to see what I mean.

An agent rescues the enquiries that slip away. It does not fix the fact that every day something else eats an hour of your time.

Who is writing this: Maruša Vrečar. I wrote sales copy for more than twenty markets. Today I build AI systems that take over the work small businesses never have time for: writing quotes and service descriptions, replying to customers even at 10 pm, taking bookings, organising and filing documents, competitor and price analyses, finding potential customers, and monthly reports from data scattered across five tools.

And that is just a sample — the list is not fixed. Book a free call, tell me what eats your time, and I will put a solution together for you. The quote is non-binding and the call costs you nothing. Your problem = my challenge. 🙂

If you want to go deep straight away, the process review for 290 € shows you within three days what a system could take over — and what is not worth it. I deduct the review fee from the project if you go ahead.

Book a free 15-min call

Would you rather someone set this up for you?

Fifteen minutes on a call. You tell me where your time leaks away — you get one concrete idea for your case, even if you decide not to work with me.

Maruša's AI assistant
AI assistant · chats are stored to improve it · for a quote, email Maruša
Free 15-min call

Cookies: this site uses strictly necessary cookies only. With your consent, I may in future also use analytics cookies. More in the Terms and privacy.