What is an AI agent? It's software you give a job to, and it works through the steps itself: it reads, uses tools and checks its own work. In an office, that means reading a supplier invoice email and preparing the bill in your accounting software.
So what is an AI agent, in one minute?
What is an AI agent? It's software that gets a job, decides the steps, does them with the tools it's been given, and checks the result before it stops.
The word "agent" means someone who acts on your behalf. So if you ask what is an agent in AI, the plain answer is a program that acts, not just replies. A chat window answers the question you typed. An agent is handed a job, like "enter this supplier invoice", and keeps going until it's done or it hits something it shouldn't decide alone.
You'll also see "agentic AI". A simple agentic AI definition is this: AI that takes several steps toward a goal, using tools on the way, instead of giving one answer and waiting. The name is new, but office software has followed rules for decades. What's new is that AI can now read messy things, like a rushed email or a scanned page.
We'll use one example all the way through: a supplier invoice that arrives by email. Nearly every business gets them, and the steps are easy to follow.
How is an agent different from a chatbot?
A chatbot answers what you ask. An agent is given a job and carries out the steps.
Both can run on the same kind of AI underneath, which is why people mix them up. The difference is what surrounds it. A chatbot has a text box and a reply. An agent has something that starts it, tools to read and write in real systems, a record of what it did, and rules for when to stop.
Say you run a small building firm. A chatbot can help you write a reply to a supplier, but you still copy it into your mailbox, attach the file and send it. An agent can notice the supplier's email, read the attachment, prepare the draft and put it where you'll find it.
| Question | Chatbot | AI agent |
|---|---|---|
| What starts it? | You type a question | Something happens: an email arrives, a file lands |
| What does it work with? | The text you give it | Tools that read and write in your mailbox, folders and accounts |
| How many steps? | One reply per question | Several, chosen to reach the goal |
| How does it know it's done? | It doesn't, you decide | It checks against rules it was given |
| What about unclear cases? | It gives its best guess anyway | A well-built one stops and asks a person |
| What is left behind? | A chat history | A draft in the right place and a record of each step |
What are the parts of an agent?
Every useful agent has four parts: a goal, tools, memory and checks. It's only as good as the weakest one.
Knowing the parts is the most practical way to answer what are AI agents, because they're what you'll set up and check.
The goal is the job written down, with a clear start and finish. "Help with accounts" isn't a goal. "Turn each supplier invoice in this mailbox into a draft bill with the PDF attached" is.
Tools are what the agent can touch: the mailbox, a PDF, the supplier list, the accounting software. Give it only what the job needs. An agent that enters bills has no reason to make payments, so it can't.
Memory is what it knows beyond the page in front of it, like known suppliers and invoice numbers already entered. Checks decide whether a result is good enough to hand over. Here's the invoice example.
- Goal: each supplier invoice becomes a draft bill with its PDF attached, filed under the supplier.
- Tools: read the mailbox and a shared folder, read PDFs and scans, look up suppliers, create a draft bill, file a document, send a notification.
- Memory: the supplier list, the layouts of regular suppliers' invoices, and invoice numbers already entered.
- Checks: the supplier is known, the amounts add up, the number isn't a duplicate, and it's clearly an invoice or a credit note.
- Limits: no payments, no new suppliers without the owner's yes, and no guessing on anything unclear.
Could an agent take this job through every step?
Pick one job, like entering a supplier invoice, and walk it through the steps an agent would take. Tick each line that's true today.
What does one invoice look like to an agent?
In an office, an agent helps most on a job that comes often, follows rules and ends in a document, an entry or an email.
Picture a small contractor. On Tuesday morning a building-supplies wholesaler emails a PDF invoice to the accounts mailbox. Nobody forwards it or opens it for the agent to start.
- Read: the agent notices the new email and opens the attachment. Is it an invoice, a credit note or a statement?
- Extract: it reads the supplier, date, number, amount, tax and total.
- Check: is the supplier known, do the figures add up, and has the number been entered before?
- Draft: it creates a draft bill with the PDF attached and files a copy in the supplier's folder.
- Hand over: it sends one notification saying what it did.
- Stop on doubt: if a check fails, like an unknown supplier or a refund, it creates nothing and passes the document to a person.
Where do agents go wrong?
Agents go wrong when the goal is vague, the tools are too broad, or they're allowed to guess instead of stopping.
AI is very good at producing something that sounds right. That helps with messy input and hurts when the answer must be exact. An agent that reads a blurred total and fills in a likely number has made a mistake that looks correct, which is the most expensive kind. The fix is a check that refuses to go on when the figures don't add up, so a person gets the document instead.
Broad tools turn small mistakes into big ones. An agent that can send emails can send the wrong one to a customer. One that can only create drafts can't. Limiting the tools is the simplest safety step, and it costs nothing, since you were going to look at the draft anyway.
Last, agents drift when nobody can see what they did. With a record of each step, "why was this coded here?" has an answer.
How do you keep a person in charge?
The safest way to use AI agents at work is to start with your approval on every step and move steps to automatic as they prove themselves.
That's how ours work. We call them AI employees. Each is an agent with one written job. They're rented monthly and run on our servers against your own accounts: today that means Google mailboxes, Drive, Sheets and Xero. Every action goes in a log, anything the agent can't do exactly goes to a person, and one command stops any of them at once.
Several of these jobs run today for a client: supplier bills, sales invoicing, purchase orders, invoice filing, a daily mail digest and online contract signing. Others are built to order, such as drafted customer replies, expense receipts and meeting minutes. We say which is which, because it matters when you decide what to hand over.
What an agent doesn't do is take responsibility. It doesn't decide what's owed or settle a dispute. Those stay with you, and a good agent makes your part quicker by putting a checked draft in front of you instead of a pile of unread emails.
Questions
What are AI agents in simple words?
Programs you give a job to. They work through the steps, using tools like your mailbox or accounting software, instead of answering one question at a time.
What are the seven types of AI agents?
Textbooks usually list simple reflex, model-based, goal-based, utility-based and learning agents, and some add hierarchical and multi-agent systems. In an office, the type matters less than the job, the tools it's allowed and the checks it runs.
Is ChatGPT an AI agent?
A chat assistant on its own answers prompts. It becomes part of an agent when it's connected to tools, given a job and allowed to take several steps with checks around it.
How can I use AI agents in my business?
Start with one job that repeats often and follows rules, like entering supplier bills. Give the agent only the tools that job needs, check each result at first, and move a step to automatic once it has proven itself.
Can an AI agent send emails or pay bills by itself?
It can be built to, but ours don't pay bills. Payments stay with you, and at the start you press send on any email.
Tell us your workflow. We will build it around you.
Tell us one job and how many hours it takes you each month. Write to hello@thenewaiemployee.com and we'll say plainly whether an agent fits it.
Talk to us Related service: Services


