Explainer
What is an AI agent? A plain answer for business owners
The short answer
An AI agent is software that carries out a multi-step job on its own: it decides what to do next, uses your other tools, and finishes without being asked each time. A chatbot answers a question. An automation follows a fixed rule. An agent works out the steps. Only about 9% of businesses use AI this way today.
Three things that get called the same thing
The word agent is doing a lot of unearned work in sales material, so it is worth separating three genuinely different products that all get described the same way.
An automation follows a rule you wrote. When a job is marked complete, send this text. It never decides anything and it never surprises you, which is both its limitation and its main virtue.
A chatbot answers a question you asked. It is excellent and it stops the moment you stop typing. Nothing happens in your business as a result unless you go and do it.
An agent is given an outcome rather than a step. Book this caller in. It then works out what to do: check the calendar, offer times, write the appointment, send the confirmation. It decides the sequence, and it can get the sequence wrong.
Worked example: the missed call
Somebody calls at 9:40am while you are under a sink. Compare the three.
The automation texts them: sorry we missed you. Useful, cheap, and the job still needs a human to finish.
The chatbot does nothing, because nobody asked it anything.
The agent answers the phone, works out the caller wants a boiler serviced, checks which slots are free, offers Thursday morning, books it, writes the job into your system and texts a confirmation. You find out when you look at your calendar.
Worked example: the unpaid invoice
An invoice goes thirty days overdue. The automation sends a reminder on a schedule regardless of what has happened, which is why customers get chased for invoices they already paid.
An agent checks whether payment arrived, reads the last exchange with that customer, notices they promised to pay on Friday, waits until Monday, then writes a chase that references what they actually said.
That difference is the whole category. Not doing the thing faster, doing it with the context a person would have had.
Worked example: the review reply
A three-star review arrives mentioning a specific technician and a specific delay. An automation cannot reply to it, because there is no rule that fits.
An agent reads the review, pulls up the job it refers to, sees the van broke down that morning, drafts a reply that acknowledges the delay honestly and offers to sort it, and holds it for you to approve.
The approval step is the important part, and the good products keep it. An agent that publishes to your Google profile unsupervised is a liability rather than a feature.
An agent that publishes unsupervised is a liability, not a feature.
Why almost nobody is doing this yet
Adoption of chat is near-universal in some fields and adoption of agents is not. In real estate, where 68% of agents use AI, only about 9% have it wired into their systems where it can take multi-step actions. Three quarters are still pasting text into a chat window.
The reason is not stubbornness. An agent needs access to your calendar, your customer records and your phone, and every one of those connections is a decision about trust and a piece of setup work. A chatbot needs a browser tab.
That gap is also the opportunity, in the sense that the businesses doing this properly currently have very little competition.
What to ask before buying one
Most products sold as agents are automations with better marketing, and that is fine as long as you know which you are buying.
Four questions that separate them
- What does it do when something unexpected happens, rather than on the demo path
- What can it change on its own, and what does it hand back for approval
- Which of my systems does it connect to, and who does that connection work
- What does it cost when it is busy, since per-action pricing rises exactly when the business is going well
Where to start
Not with an agent. Start with the automation, because it is cheap, predictable and fixes the most expensive gap most small businesses have, which is nobody answering.
Once that is running you will know from real data where the remaining work actually is. That is a much better basis for choosing an agent than a demo, and it means the first one you buy solves a problem you have measured rather than one you were sold.
Common questions
- What is the difference between an AI agent and a chatbot?
- A chatbot answers a question and stops. An agent is given an outcome and works out the steps to reach it, using your other systems. The chatbot ends with text; the agent ends with something changed in your business.
- Is an AI agent the same as automation?
- No. An automation follows a rule you wrote and never decides anything. An agent decides the sequence itself, which makes it more capable and also means it can get the sequence wrong.
- Are AI agents safe to let loose on customers?
- Only with an approval step for anything public or irreversible. Booking an appointment is low risk. Publishing a reply to a review, issuing a refund or sending a quote should be drafted and held for a person.
- How many businesses actually use AI agents?
- Very few. In real estate, where 68% of agents use AI in some form, only around 9% have it integrated where it can take multi-step actions. Three quarters still paste text into a chat window.
- What should I automate before buying an agent?
- Whatever stops you losing work you already have: answering the phone, replying to missed calls, sending reminders. Those are cheap and predictable, and running them tells you where an agent would actually earn its money.