What an AI Agent Actually Is

You have probably seen the phrase everywhere this year. Your software has started offering "agents". Headlines say they will run whole departments. Someone on LinkedIn has "hired" twelve of them.

Most of that is noise. Underneath it is one genuinely useful idea, and it is simple enough to explain in a sentence.

The one-sentence version

An AI agent is AI that you give a goal, and it works out the steps and carries them out itself, using tools you have allowed it to use.

That is the whole idea. Everything else in this course is about using it well.

Three things that look alike but are not

The easiest way to understand an agent is to put it next to the two things people confuse it with.

A chatbot answers

ChatGPT, Claude or Grok in their everyday form are chatbots. You ask, they answer. You ask again, they answer again. Nothing happens in the world unless you take the answer and do something with it yourself.

You: "Write a polite reply to this customer asking to move their appointment." Chatbot: writes the reply. You copy it, open your email, paste it, and press send.

An automation follows your steps

An automation is a fixed set of steps you design once, and a tool repeats them exactly, every time. "When a form is submitted, add the person to my mailing list and send them the welcome email." It is reliable precisely because it never thinks. If anything unexpected happens, it either stops or does the wrong thing confidently.

An agent works out the steps

An agent sits between the two. You give it a goal rather than a list of steps, and it decides what to do next, does it, looks at what happened, and decides again.

You: "Find the customer's booking, move it to the first free slot next week, and draft a reply confirming the new time. Show me the reply before it goes." Agent: opens your calendar, finds the booking, looks for free slots, moves it, writes the reply, and stops to show you.

Same job. The chatbot gave you words. The automation could only have done it if you had predicted this exact situation in advance. The agent did the work, and stopped where you told it to.

The four parts of every agent

Every agent, whichever company makes it, is built from the same four parts. Once you can see them, the marketing becomes much easier to read.

1. The brain. The AI model that does the thinking, such as the one behind ChatGPT, Claude or Grok. It reads the goal, makes a plan and decides each next step.
2. The tools. What the agent is allowed to touch: a web browser, your email, your calendar, a spreadsheet, a file folder. No tools, no agent. A model with no tools can only talk.
3. The instructions. What you have told it: the goal, the rules, what it must never do, and when to stop and ask you. This is the part you control most, and chapter 7 is all about it.
4. The memory. What it can remember while it works, and sometimes between jobs: your preferences, what it did last time, the notes you gave it.

When an agent disappoints, one of those four is nearly always the reason. It had the wrong tools, unclear instructions, or no memory of what mattered.

The loop it runs

What makes an agent an agent is that it runs a loop rather than a single step:

1. Understand the goal. What am I being asked to achieve?
2. Plan. What is the first thing to do?
3. Act. Use a tool: search, open, read, write, click.
4. Check. Did that work? What do I know now?
5. Repeat or stop. Go round again, or stop because the job is done, or stop because it has reached something it was told to ask about.

A chatbot does step 1 and then hands everything back to you. An agent keeps going round. That is where the time saving comes from. It is also where the risk comes from, because every trip round the loop is a decision you did not personally make.

Why this matters now

Agents are not new as an idea. What changed over the last two years is that the brains got good enough to plan sensibly, and the big AI companies started building tools into their products. So an agent is no longer something you need a developer to build. It is a mode inside tools you may already pay for.

For a small business, that is the real story. A lot of your week is made of jobs that are too varied to automate and too dull to deserve your full attention: chasing, checking, comparing, filing, drafting, researching. Those jobs are exactly the ones agents are starting to be good at.

Why they still need you

An agent is not an employee. It has no judgement about your business beyond what you give it, no stake in whether a customer is happy, and no sense that something "feels off". It can be fooled by what it reads on a web page. It can be confidently wrong, just like the chatbot it is built on, except that now it can act on being wrong.

That is why this course spends as much time on choosing the right jobs and setting limits as it does on the tools themselves. The owners who do well with agents are not the ones with the cleverest setup. They are the ones who were clear about what to hand over, and kept the right decisions for themselves.

An example: Maya's Monday morning

Maya runs Hartley Heating, a six-person heating and plumbing firm. Every Monday she spends about 90 minutes on the same kind of work. She checks which weekend enquiries came in, looks up whether each postcode is in their area, checks the diary for the earliest visit, and replies to each one.

• With a chatbot, she could get help writing each reply. She still does all the looking up.
• With an automation, she could send an instant "thanks, we'll be in touch". It cannot check the diary or judge the postcode.
• With an agent, she could say: "Go through this weekend's enquiries. For each one, check the postcode against our area list, find the earliest free survey slot, and draft a reply offering it. Don't send anything. Put the drafts in a list for me to approve."

The agent does the looking up. Maya does the approving. Her 90 minutes becomes 15.

We will follow Maya through the rest of the course.

If you are using this for yourself, the same idea applies at home: "Compare these three energy tariffs against my last year of usage and tell me which is cheapest, with your working" is a job for an agent, not a chatbot.

Common mistakes

• Calling any AI feature an "agent". If it cannot act with tools, it is a chatbot with a new label. Ask: what can it actually touch?
• Handing over a goal with no stopping point. An agent without "stop and show me before you…" will carry on until it thinks it has finished. That may be further than you wanted.
• Expecting it to know your business. It knows what you tell it. The rest is guessing.

Quick check

1. Your software offers an "AI agent" that writes social posts when you type a topic. It cannot publish, schedule or look at anything. Is it an agent?
2. You set up "every time an invoice arrives in my inbox, save the attachment to the Invoices folder". Chatbot, automation or agent?
3. An agent was asked to "tidy up the shared drive" and deleted files you needed. Which of the four parts most needed fixing?

Answers. (1) No. It is a chatbot. It has no tools and cannot act. (2) An automation: fixed steps, repeated exactly. (3) The instructions. It needed limits ("move, never delete") and a checkpoint ("show me the plan before you change anything"). Tighter tools would help too.

Key takeaways

• An agent is given a goal, and works out and carries out the steps itself.
• Chatbots answer; automations repeat; agents decide and act.
• Every agent has a brain, tools, instructions and memory. Problems nearly always trace back to one of them.
• The loop that saves you time is the same loop that creates risk, so you decide where it stops.

Action Steps

1. Open your workbook at Section 1 and write, in your own words, one sentence explaining what an AI agent is, as if to a member of your team.
2. List three jobs from the last week that you did by "looking things up and then doing something". These are your first agent candidates. We will test them in chapter 3.
3. Check the AI tools you already pay for. Does any of them offer an agent, work or task mode? Note it down. You may already own what you need.

Next: now that you know what an agent is, chapter 2 shows what agents are actually doing for small businesses today, and where they are not worth the trouble.