What an AI agent is, and what it can do in a small business

What an AI agent is, in plain words

At the heart of every AI agent is a language model, the same kind of technology that sits behind ChatGPT. On its own, the model only writes text. It becomes an agent when you give it a goal and access to tools: reading emails, searching a spreadsheet, filling in a document, adding something to the calendar.

Then it works in a loop. It looks at the task, picks a step, carries it out, sees the result and decides what comes next. For example: "Go through today's enquiries, sort them by urgency and write a draft reply to each." The agent opens the inbox, reads the emails, checks the free slots and leaves the drafts for you.

The key words are "on its own". Nobody wrote every step in advance. That is both the strength and the risk of an agent.

Chatbot, automation and AI agent: the difference

The three are often confused, because all three "answer by themselves". The difference is who decides what happens.

• Chatbot It talks. It answers customer questions on the website, in Viber or in Messenger, usually from information you have given it. A chatbot usually talks, but does not do work in your other systems.
• Automation It follows steps someone described in advance: when an enquiry arrives, add it to a spreadsheet and send an email. It does the same thing every time, quickly and predictably. It cannot handle what nobody foresaw.
• AI agent It receives a goal and chooses the steps itself. It copes with messy emails, different formats and unexpected cases. That is also why it can go wrong in ways nobody foresaw.

What you can trust an agent with, and what you cannot

An AI agent can sound confident and still be wrong. It knows nothing about your business beyond what you show it. So the rule is simple: give it work a person can check quickly, not decisions that cannot be undone.

• Good at Sorting and summarising emails and enquiries, drafting replies and quotes, pulling data out of invoices and documents, searching your own documents, preparing reports for review.
• Only with a person's approval Anything that goes out: an email to a customer, a post, a change to an order or a schedule. The agent prepares, a person presses "send".
• Do not trust it with Payments and money, deleting data, legal, medical or tax advice, and final decisions about people, such as hiring or turning a customer away.

Examples in a small business

• The inbox The agent reads new emails, separates enquiries from advertising, sorts them by urgency and leaves a draft reply. The owner reviews and sends.
• Documents Incoming invoices and contracts are read, the data goes into a spreadsheet, and anything missing is flagged. The accountant checks before anything is booked.
• Quotes From the customer's description the agent puts together a draft quote using your template and your prices. A person reviews and sends it.
• Questions from the team Staff ask the agent where a procedure is written down or what the contract with a supplier says, instead of interrupting the owner. The agent answers only from your documents and says where the answer came from.
• More examples Not every one of these tasks needs an agent. Many are handled more reliably by an ordinary automation.

How to start safely

• One narrow job Not "an agent for everything", but one piece of work with a clear start and end, such as sorting enquiries.
• A person approves The agent prepares, a person decides. Nothing goes to a customer, a supplier or a bank without confirmation.
• Minimum access Give it only what the job needs. First the right to read, later the right to write, if it turns out to be necessary.
• A record of every step You need to see what the agent read, what it decided and why. Without that record you cannot find a mistake.
• A trial on past cases Run it on enquiries or documents you have already handled, and compare its result with what you did.
• Questions for the provider Where is your data processed? Is it used to train the model? What happens if you stop the service? Get the answers in writing.

How we work with AI agents

We build our own software with Ledger, our own AI build engine that writes and checks code. People decide what gets built and accept the finished work. We apply the same rule for clients: the agent does the preparation, a person approves.

If you are not sure whether you need an agent, an automation or neither, the free 5-minute check shows where your business loses time and money, using your own numbers. The €390 audit maps your processes and ranks what is worth building first.