What is an AI agent? A plain-English guide for business owners
"AI agent" is everywhere at the moment, usually wrapped in jargon. Here is what it actually means for a business like yours, and what it does not.
The short answer
An AI agent is software that can do a job, not just answer a question. You give it a goal and some rules, connect it to the tools your business already uses, and it works through the steps on its own: reading, deciding, doing, and checking back with a person when something is unusual.
A chatbot waits for you to type something and replies. An agent notices that a supplier invoice has landed in your inbox, reads it, checks it, enters it into Xero and only bothers you if something looks wrong.
What an agent is made of
- A trigger. Something that starts the work: an email arrives, a form is submitted, it is 7am on Monday.
- A brain. A large language model (the same kind of AI behind ChatGPT or Claude) that can read messy information and decide what to do with it.
- Hands. Connections to your systems, such as Gmail, Outlook, Xero, your CRM or your booking tool, so it can actually take action.
- Rules and a supervisor. Your business rules, plus a clear point where it hands over to a person.
Everyday examples
Most of the useful agents in small and mid-sized businesses are not futuristic. They do boring, repetitive work reliably:
- An inbox assistant that sorts email, drafts replies in your voice and escalates the few that need you.
- An enquiry handler that replies to every new lead within minutes and books good ones into your calendar.
- A supplier invoice agent that reads bills from email and enters them into Xero or MYOB.
- A Monday report that pulls numbers from four systems and writes a plain-English summary of the week.
What agents are not good at (yet)
Agents are excellent at repetitive work with clear rules and lots of reading and writing. They are not a replacement for judgement, relationships or anything where a wrong answer is expensive and hard to undo. The best setups keep a person in the loop for exactly those moments.
That is also why so many AI projects disappoint. A widely reported 2025 MIT study found most business AI pilots made no measurable difference to the bottom line. The common thread is that nobody owns the result: a tool gets bought, nobody connects it to how the business actually runs, and it quietly dies.
How to tell if a job suits an agent
Ask three questions about any task in your business:
- Does it happen often, at least weekly?
- Could you write down the rules for doing it well?
- Does it involve reading, writing, copying or checking information between systems?
If the answer is yes to all three, it is very likely a job an agent can do. Start with one, prove it saves real time, then add the next.