What is an AI agent, in plain English?

An AI agent is software that finishes a job, not just answers a question. Give it a goal, like "reply to this enquiry and book the call", and it works through the steps: reads the email, drafts the reply, checks the diary, sends the invite. Useful for repetitive multi-step work, oversold for everything else.

Last updated 11 June 2026

You keep seeing the word everywhere. Every tool you already pay for has announced an "agent", every LinkedIn post says agents will run your business by Christmas, and nobody has paused to explain what one is. The honest definition is short.

ChatGPT answers a question and stops. An agent keeps going. It takes a goal, breaks it into steps, uses other tools to do those steps, looks at the result, and carries on until the job is done or it gets stuck. The pieces are the same AI you already know; the difference is that it has hands as well as a mouth. It can read your inbox, write to your calendar, update your spreadsheet, send the email.

What that looks like in a real business

Forget the demos of agents booking flights. The versions that earn their keep in a small business are mundane:

  • An enquiry lands on your website. The agent reads it, looks up what the prospect's company does, drafts a reply in your voice, books a slot in your diary and logs the lead. You see a finished result, not a to-do.
  • An invoice goes overdue. The agent checks whether payment arrived, picks the right chasing tone for a first or third reminder, sends it, and flags the account if a customer goes quiet.
  • Every Monday it pulls your sales numbers, bank feed and ad spend into one summary before you have made coffee.

Notice what those have in common: a clear trigger, defined steps, output a human can check. That is the territory where agents work today. The same selection rules as any first automation apply.

Where the word is being abused

Half the products calling themselves agents are a chatbot with a new badge. I build both, and the test I use is simple: does it finish the job without you, or does it hand you a draft and call that autonomy?

The other abuse is scope. An agent given a vague goal and broad permissions will do vague things with broad confidence. The failures you read about, agents deleting data or emailing the wrong customers, are nearly always a scoping failure: someone gave software an open-ended brief no employee would get in their first week.

Do you need one?

You need one if you have a repetitive, multi-step process with clear rules and real volume. Enquiry handling and lead follow-up are the classic cases because every hour of delay costs money.

You do not need one to start with AI. Most businesses get their first wins from far simpler things: a well-set-up assistant for drafting, a single automation with one trigger and one output. Buy the outcome, not the word. If a salesperson cannot tell you exactly what jobs their agent completes, which systems it touches and what happens when it is unsure, the agent is the demo, and you are the test.

Answered by Dean Cookson, Founder and CEO at Operosus.

Now try it yourself

Most answers on this page come with a do-it-yourself route, and we would honestly rather you took it. Get stuck and the weekly Cook-a-Long is free. Decide it is not worth your hours, and that is what we are for.

Book a call. 30 minutes, no pitch deck.