“AI agent” is the buzzword of the moment and, as always, it's being stuck onto very different things. Onto a bog-standard chatbot with a new name. Onto the same automation we've had for years. And, every now and then, onto something that genuinely is new.
The difference fits in one sentence: a chatbot answers, an agent acts. If something has changed by the end of the conversation —an appointment in your calendar, an order logged, an email sent— it was an agent. If it only replied to you, it was a chatbot.
What an AI agent is, without the hype
A program that is given a goal, decides for itself which steps are needed and carries them out using real tools: your calendar, your CRM, your database, your email. It doesn't follow a fixed script with its branches drawn in advance, which is what a classic chatbot did.
That's what makes it useful, and also what makes limits necessary. A program that decides is a program that can decide badly.
A chatbot answers. An agent acts.
What it genuinely does for a small business
What we see working today, not what gets shown in the videos:
- ✓Answering out of hours. Enquiries don't wait for office hours: they arrive at night, at weekends and in August. An agent takes the details, resolves what it can and leaves the rest ready for the next morning, instead of letting it get lost.
- ✓Booking appointments. It looks at the real gaps in the calendar, proposes, confirms and sends reminders. This is where everything used to fall apart: the bot said “we'll call you” and nobody called.
- ✓Filtering and qualifying. Before a salesperson spends an hour on it, the agent has already asked about budget, timescale and exactly what's needed.
- ✓Searching what you already know. Prices, delivery times, terms. The information is already in the business; what isn't there is someone free to go and look it up every time.
Where the agent lives
The same agent can handle several channels, and that's a good part of the point. The three usual ones:
WhatsApp. Where your customer already is. Nothing new to teach them.
The website. Useful above all when the visitor arrives from a campaign and you want to answer their question before they leave.
The phone. The one that surprises people most and works best in businesses with a lot of repetitive calls: opening hours, availability, the status of an order.
What to decide before switching it on
This is where a project that goes well parts company with one that ends in tears. Four decisions, and none of them technical:
What it can do on its own and what it can't. Book an appointment, yes. Apply a discount, probably not. Write it down before building anything.
When it has to hand over to a person. An agent that keeps pushing when it doesn't know does more harm than one that says “Marta will look into this and call you in ten minutes”.
What data it touches. If it's going to read or write customer data, that goes into your record of processing activities and has to be stated in your privacy notice. It isn't optional.
Who reviews it. You read every single conversation for the first week. It's the only way to find out what your customers actually ask, which is almost never what you thought.
No hype: an agent doesn't replace anyone on your team. It takes the repetitive part off their hands —the same question forty times— so that person is free when the conversation gets interesting. Anyone selling it to you as “lay off half your team” has never built one.
Where to start without spending a fortune
Take a single task, the most repetitive one you have, and build it properly. Not the assistant that knows everything: the question you get asked forty times a week.
Measure it with two numbers, not a gut feeling: how many conversations get resolved without anyone stepping in, and how many end in something (an appointment, an order, a useful contact). That tells you within three weeks whether to expand it or switch it off.
And if the problem you have isn't answering enquiries but the paperwork that comes afterwards, then what you're looking for isn't an agent but process automation: they're different things and they're solved differently.
Frequently asked questions
What's the difference between a chatbot and an AI agent?
A chatbot answers by following a script written in advance; an agent is given a goal, works out which steps are needed and carries them out using real tools, such as your calendar or your CRM. The practical test: if something has changed by the end of the conversation —an appointment booked, an order logged— it was an agent.
How long does it take to set up an AI agent?
It depends on the scope, which is exactly why it pays to start with a single task. An agent that handles your most repeated question and books appointments takes days to build, not months. It's the assistant that knows everything that drags on for ever and almost never ends up working.
Can an AI agent replace someone on my team?
No, and anyone selling it that way has never built one. What it does is take the repetitive part off their plate —the same question forty times over— so that person is free when the conversation gets interesting and there's something to win.
What happens to my customers' data?
If the agent reads or writes personal data, that processing goes into your record of processing activities and has to be reflected in your privacy policy, just like any other tool. It isn't optional paperwork: it's the part you decide before switching anything on.
How do I know whether it's working?
With two numbers, not a gut feeling: how many conversations are resolved without a person stepping in, and how many end in something concrete (an appointment, an order, a useful contact). That's enough to tell you within three weeks whether to expand or switch it off.
Shall we build you one and test it?
We pick the most repeated task with you, build the agent with clear limits and measure whether it works. If it adds nothing in three weeks, we switch it off.