What is an AI agent?

AI agents in plain terms

Imagine a new colleague who never walks away from unfinished work. They get a clearly described task – say, “enter every incoming order from email into the system” – and do it with the same care every time, even at night and at weekends. When they are not sure, they do not decide alone; they ask a person.

That is exactly how an AI agent works. It is a program built on a large language model (the same technology behind ChatGPT), but it has three things an ordinary chat does not:

• A goal – it knows what result it has to deliver, not just what to reply to.
• Tools – it can read emails, look into spreadsheets, write to systems or send messages.
• Rules – it knows what it may do on its own and when it must hand the work to a person.

How an agent works, step by step

Take a common example from an accounting firm: a client emails an invoice.

1. It picks up the trigger – the agent notices a new email with an attachment.
2. It understands the content – it reads the invoice and pulls out the supplier, amount, due date and account number.
3. It checks the details – it compares them with what is already in the system and flags anything that does not match.
4. It prepares the output – it pre-fills the entry in the accounting software.
5. It hands over to a person – the accountant checks and confirms at a glance. Unclear cases come with an explanation of why they are unclear.

Routine that used to mean retyping becomes checking. The decision stays with a person – this principle is called human-in-the-loop.

What an AI agent is not

• It is not a website chatbot. A chatbot answers questions; an agent finishes tasks. We explain the difference in detail in AI agent vs. chatbot.
• It is not a robot that replaces your team. The agent takes over repetitive routine, so people have more time for work that needs judgement and customer relationships.
• It is not infallible. Language models can get things wrong (so-called hallucinations). That is why an agent is built to work from your verified documents, with a person approving the important steps.

Where an agent makes sense

It works best for work that repeats, has a clear input and output and can be checked. Typically:

• sorting incoming mail and enquiries,
• copying details from emails and documents into systems,
• answering repeat questions based on your documents,
• tracking deadlines and chasing missing paperwork,
• preparing documents, overviews and draft texts.

You will find specific tasks for 23 industries in the agent catalogue. For advice on choosing the first one, see How to choose your first task for AI.

Frequently asked questions