What AI agents are, and what they can actually do for your business
“AI agent” has been used for almost everything over the past year, which has left it sounding either magical or suspect. In practice it means something quite specific, and it is worth separating from the chatbot you may already be considering.
The difference from a chatbot, in one line
A chatbot answers. An agent does.
A chatbot takes a question and returns an answer. An agent takes a goal, breaks it into steps, uses your tools to carry them out, and reports what happened.
An everyday example from an online store:
- Chatbot: “Where is my order?” → reads the status and replies.
- Agent: takes the day’s emails, spots which ones are return requests, opens a ticket in your system, asks for the missing details and updates the customer — without anyone asking it to.
What it actually takes over
The rule is simple: repetitive flows with clear steps and a clear “right answer”. Among the small and mid-sized businesses we work with, the most common are:
- Entering data from emails, PDFs or invoices into an ERP or accounting system
- Research and short written summaries (competitors, prices, suppliers)
- Triaging an inbox: categorising, prioritising, routing to the right person
- Preparing quotes from a template and the customer’s details
- Reconciling files — finding what does not match between two systems
- Updating the CRM after every interaction
What it does not do well: decisions needing judgement, negotiation, or knowledge that exists only in someone’s head. If every case is different, the agent will keep asking you, and you will not save any time.
“What if it gets something wrong?”
That is the right question, and the answer is not “it won’t”. It will. How you handle that is a design decision, not a technology one:
Boundaries. The agent can reach only what that specific flow requires. You do not hand over the keys to the whole business because you can.
Approval points. Where the risk is real — money leaving, an email going to a customer, something being deleted — it stops and asks a person. Everywhere else it proceeds on its own.
A record. Every action is logged: what it did, when, with what data. When something goes wrong you can see exactly where, instead of guessing.
In practice we almost always start with the agent in a “it prepares, a human approves” role. Once you have watched it be right fifty times in a row, you remove the approval step wherever you are comfortable.
What it costs
There is no price list, because the cost is not the agent — it is the connections. An agent working only inside your inbox is a small project. One that has to talk to an ERP, an e-shop and your accounting system at once is a bigger one.
The calculation that matters runs the other way: how many hours a month go into this flow today? If it is four, it is probably not worth automating. If it is forty, the agent pays for itself quickly and you know that from the start.
Where to start
Do not look for the most impressive flow. Look for the boring one:
- Something a person does every day or every week
- With steps you could write down on paper
- Where the “right result” is obvious
- And where, if it goes wrong, you notice before any damage is done
That flow will give you a real number within weeks. With that number you decide whether to go further — rather than deciding on the strength of a demo.
Have a flow in mind? Book a free 30-minute call and we will tell you honestly whether it is worth automating. See also what we do with AI agents.