Chatbot, agent or AI assistant: how they differ and which one your business needs

What most people call a chatbot is a rule-based system. The client types something or taps a button, and the program answers from a script someone wrote beforehand: if they say this, reply that. It works like a tree of options. "Tap 1 for opening hours, 2 for a quote, 3 to speak to a person."

Within its own ground it does the job well. If the questions you get are always the same and fit into four or five buttons, a chatbot like this covers them easily, it's cheap, and it never makes anything up because it can only say what's already written.

The trouble shows up when the client steps off the script. They ask something in their own words, fold two doubts into one sentence, or raise a case nobody foresaw, and then the chatbot goes blank or sends them back to the start. IBM explains it in its guide on chatbot types: because you can't leave an answer programmed for every possible question, the rule-based bot has little flexibility and ends up leaving the client feeling like they're talking to a wall.

An artificial-intelligence assistant starts from a different place. Instead of following a tree of buttons, it reads the client's sentence and understands what they're asking, even if they write it their own way, with typos or mixing topics. The industry calls it natural language processing, and for the client it comes down to something simple: they can write the way they'd speak to a person.

That difference changes what the tool can do for you. A well-built assistant knows your business, from your services to your prices and the area you cover. It answers a specific question without forcing the client through a menu, takes down the details of a job and, when things get complicated or the client asks, says it's handing over to a person. It leaves nobody stranded.

When understanding is joined by doing (checking a calendar, booking an appointment, handing off to a human at the right moment), that's what many now call an agent. For you the label doesn't matter. What matters is that the client walks away served and you find out what happened.

The boundary has little to do with whether the tool is modern or old. It comes down to a practical question: do your clients always ask the same thing, or is every conversation different?

If your business answers four repeated queries (hours, address, a fixed rate, little more), a menu-based chatbot sorts it out without spending more than you need. That's a perfectly valid answer and, for plenty of people, the most sensible one.

If instead every client turns up with their own case, haggles, describes an odd fault or needs someone to understand them before deciding, a menu falls short and ends up frustrating. There, an assistant that understands and gathers the information properly saves you missed calls and quotes that go cold. The question worth asking isn't which tool is better in the abstract, but which fits the way people write to you.

An assistant that understands and speaks fluently comes with a flip side. Without limits, it can answer things that in your trade are better left unanswered, or let slip a detail it shouldn't. In a service business that matters a great deal.

That's why, when I build a tool, the safety that belongs to the trade goes in from day one. The first case was a locksmith: his assistant serves clients while his hands are on the lock, and it never explains how to open one. Every trade has its own red lines, and the tool has to know them just as the professional does.

There's also something that will soon be required. From 2 August 2026, Article 50 of the EU AI Act applies: if a client is talking to an automated assistant, they must be told they're talking to an AI and not a person. A "hi, I'm the assistant for [your business]" at the start of the chat covers it nicely and, along the way, signals that you take things seriously.

I don't start from an assistant meant to serve everyone alike. I sit down with you, understand how you actually work, and build the tool for your trade. We test it with your clients and, when it works, there's a person behind it keeping it going with you. For many, the first version is an assistant that handles WhatsApp while they work.

Everything runs inside Europe, on Google Vertex AI with Gemini models, without going through OpenAI. I don't train any model on your conversations or your clients' data, and we sign the data processing agreement that covers you under the GDPR. The assistant always introduces itself as an AI from the first message.

If you run a service business and the idea of AI lending a hand without losing control appeals to you, start by trying it. You can talk to the real assistant for locksmiths at aistant-eight.vercel.app/demo/cerrajero, or send me a WhatsApp at +34 632 402 668 and tell me your case. If a simple chatbot is enough for you, I'll say so; and if yours calls for an assistant built around your trade, we'll set it up properly from the start.