What is an AI agent? Explained without the jargon

By [FOUNDER NAME], Founder — about Mach Lilies · Updated 4 July 2026

"Agent" is the technical word. You will see it in almost every article about AI this year, usually with no explanation. This guide explains it properly: what an agent is, how it differs from a chatbot, what one can safely do in a business like yours, and — just as important — when you don't need one at all.

No assumed knowledge. If a sentence here makes you re-read it, that's our failure, not yours.

What does an AI agent actually do?

An AI agent works through a task step by step. It reads information, decides what's needed next, does it, and checks the result. A chatbot stops when it has answered your question. An agent keeps going until the job is finished — and it writes down every step it takes.

Here's a concrete example. An accountancy practice needs bank statements from 40 clients every quarter. On Monday morning the agent checks the tracker and finds 12 clients still owe records. It emails each one, politely, from the practice's own address. By Wednesday, five have replied. The agent reads the replies, files the statements in the right client folders, and updates the tracker. One client has asked a question instead, so the agent passes that email to a member of staff with a one-line summary. On Friday it sends the manager a short report: seven clients still outstanding, gentle nudges scheduled for Monday.

Nobody on the team wrote those chase emails, filed those statements or updated that tracker. That is what an agent changes: the task happens, instead of waiting for someone to find time for it.

What's the difference between an AI agent, a chatbot and ordinary software?

A chatbot answers questions, but you still do the task yourself. Ordinary software follows one fixed rule and never varies. An AI agent sits between the two: it can read a situation, make small judgement calls, and carry out the steps itself — with a person approving anything important.

A chatbot answers you, then stops

ChatGPT and Claude are chatbots. Ask one to "write a polite reminder email" and you'll have a good draft in seconds. But you still copy it, send it, watch for the reply, file the attachment and remember to follow up next week. Close the tab and it forgets you exist. For many owners a chatbot at about £20 a month is genuinely useful — we say so plainly further down.

Ordinary software follows one rule, forever

Think of an Outlook rule that moves every email from HMRC into a folder, or a Xero bank feed that pulls in transactions overnight. This kind of software is cheap and dependable, but rigid. It does exactly what the rule says and nothing more. The moment reality varies — a client replies "which statement do you mean?" — the rule has no answer, and a person has to step in.

An agent handles the messy middle

An agent can cope with the variation. It reads the client's reply, realises it's a question rather than a bank statement, and passes it to a person instead of filing it blindly. It notices that one client always sends photos instead of PDFs, and files them correctly anyway. It works across steps — read, decide, act, check — the way a careful assistant would.

What can an AI agent safely do in a small business?

The safe jobs are the repetitive ones with a clear finish line: chasing missing documents, sorting the inbox, preparing the weekly report, checking invoices for odd amounts. A useful test: if you could write the steps on one page for a new starter, an agent can probably take the job on.

• Chasing documents and information. Politely nags clients for missing paperwork until it arrives, then files it in the right place.
• Sorting the inbox. Reads incoming email, drafts replies to the routine ones for approval, and routes the rest to the right person with a short summary.
• Preparing reports and packs. Pulls information from your systems into the weekly and monthly documents someone currently builds by hand.
• Checking invoices and admin. Spots the exceptions — the missing reference, the odd amount — and queues them for a person to look at.

And what shouldn't an agent do? Anything that needs professional judgement, or carries real risk if it goes wrong: giving clients advice, signing work off, making final decisions about money or people. Those stay with your team. The agent's job is to clear the routine work out of their way, so they have more time for the judgement calls.

For a longer list of jobs, with the hours each one typically saves, see our guide to how AI can help your small business.

Why call them "AI helpers" instead of agents?

Because the word "agent" confuses more people than it helps. In Britain, an agent sells houses or spies for a living. "AI helper" says what the software actually is: something that helps a person do their work, without replacing them. Everywhere else on this site, "helper" means exactly what this guide calls an agent.

The industry will keep inventing new words for all of this. You don't need to learn them. If a supplier can't explain what their software does in one sentence your whole team would understand, be careful.

How do you stay in control of an AI agent?

Two plain rules. First, a person approves anything important — and you decide what counts as important. Second, everything is logged: every email sent, every file moved, every decision made. Between the approvals and the log, an agent is easier to supervise than most new members of staff.

Approval, in practice. At the start, nothing goes to a client without a one-click yes from your team — it takes seconds. After a few weeks, most businesses loosen the line: routine chases go out on their own, and only the unusual ones wait for a person. You can tighten the line again at any time, for any reason.

Logging, in practice. The log is a plain record anyone can read — "emailed Mrs Patel at 9.02 asking for her March statement; filed her reply at 14.31". If something ever looks odd, you can see exactly what happened and when. You should also get a simple weekly summary of what your helper did, in plain English, without having to ask for it.

If you want the fuller picture — where your data goes, who can see it, and what agreements sit underneath — it's all on our data safety page, written so you can forward it to your IT adviser.

When is an AI agent not what you need?

Quite often. If you mainly want help with writing and answers to questions, a chatbot subscription at about £20 a month is enough. If the task follows one fixed rule, a free rule in Outlook or Xero already does it. An agent only earns its keep on messy, repetitive work that eats hours every week.

• You want better letters, emails and summaries. A ChatGPT or Claude subscription does this well. Don't pay anyone thousands of pounds for it.
• Every email from one sender should go to one folder. That's an Outlook rule. It's free and takes two minutes to set up.
• The job only happens once or twice a year. The setup cost won't pay back. Keep doing it by hand, with a chatbot to speed up the writing.
• The job is done differently every time, and no two people agree how. Fix the process first. An agent pointed at chaos just produces faster chaos.

The honest test is money. As a rule of thumb, an agent makes sense when it would save around eight staff-hours a week. We've put the full numbers — setup, monthly running costs and the payback sums — in our guide to what AI automation costs a UK small business.

Questions owners ask about AI agents