Short answer: ChatGPT is a tool that a person starts every single time. An AI agent is a system that works on its own. If you do the job a few times a week and don't mind copying the result to where it needs to go, ChatGPT is enough - and you don't need to pay for automation. The line starts where something has to happen without you.
Three questions that show the line
When someone asks me this, I don't explain the technology. I ask three questions:
- 1. Who presses the button? With ChatGPT, always a person. No person at the screen, no result. An agent starts working from an event: an incoming email, a new order, an approaching deadline.
- 2. Where does the data come from? ChatGPT can't see your inbox, CRM, stock levels or accounting. Everything it needs, you have to type in or upload yourself. An agent is connected to those systems.
- 3. Where does the result go? ChatGPT's answer stays in the chat window. Not on the invoice, not in the CRM record and not in the customer's inbox. Someone has to carry it there.
If you answered "a person, and I don't mind" to all three, you can stop reading here. Seriously.
ChatGPT shortens the thinking. Automation removes the carrying. They're two different things, and for most businesses the second one is the expensive one.
When ChatGPT really is enough
This list costs me clients, but I'd rather not lie. ChatGPT is enough when:
- you're writing, translating, shortening or polishing text;
- you're doing a one-off analysis - reviewing a contract, summarizing a document, double-checking a calculation;
- you're generating ideas, options or proposal drafts;
- the same job repeats less than a few times a week;
- a person reads and approves the result anyway, and it takes them a minute.
If that's you, I'll save you a call with me: €20 to €25 a month for a subscription and off you go. A build won't pay off for you, and I don't like projects that don't pay off.
When it's no longer enough
The line isn't about complexity, it's about repetition multiplied by a person in the loop.
A concrete example. A company receives about 60 supplier invoices a week, roughly 260 a month. With ChatGPT it goes like this: open the PDF, upload it, ask, read the answer, check the amounts, copy them into the accounting system, move on to the next one. ChatGPT genuinely helps here - it pulls the data out fast. But a person stays in the middle the whole way.
| 260 invoices a month | By hand | With ChatGPT | With an agent |
|---|---|---|---|
| Time per invoice | ~5 min | ~3 min | ~0.5 min to check |
| Person's time per month | ~22 h | ~13 h | ~2 h |
| Person's cost ($46.89/h) | $1,032 | $610 | $94 |
| Tool or agent cost | – | ~€22 subscription (about $25) | $300 (monthly minimum) |
| Total per month | $1,032 | $635 | $394 |
Calculated with a person's hour at $46.89 and our price of up to $10 per hour of AI work. The agent does about 20 hours of work here, which would be $200, so this job alone lands on the $300 monthly minimum. The gap between ChatGPT and the agent is about $240 a month, or roughly $2,900 a year. But the key thing isn't the amount, it's the proportion: ChatGPT takes roughly 40% of the person's time off their plate, the agent about nine-tenths. You can plug in your own numbers in the calculator.
And note: at 260 invoices the difference isn't dramatic yet - about three working weeks of one person's time a year. At 2,600 it becomes obvious. That's why the answer to "is ChatGPT enough?" almost always depends on volume, not on technology.
"What if I give ChatGPT access to my files?"
Fair question. File uploads, projects, custom GPTs and connectors to some services really do narrow the gap. But four things remain:
- Someone still has to start it. A chat doesn't open by itself at 7 a.m. on a Friday when an order comes in.
- No guarantee of consistency. A hundred cases have to be handled the same way. Nothing in a chat window ensures that.
- No action log. When you need to show a month later who did what and when, a chat history isn't an answer.
- No error handling. What to do when data is missing, a service doesn't respond or a case is non-standard. In automation, these boring things are most of the work - more on that separately.
Then there's the data question: what you upload and where, who sees that data, and whether you're allowed to do it at all. That's a separate topic, covered here.
The third option nobody suggests
If you're not sure which one you need, the best step isn't buying automation. The best step is doing the job by hand with ChatGPT for a few weeks.
That's the cheapest possible way to find out: if ChatGPT handles the job when you give it the right data, an agent will handle it automatically. If it doesn't, an agent won't either - it'll just cost more.
You also get two things out of it: a clear number of hours and an almost ready technical brief, because you already know how the job needs to be explained. Clients who come to us after a trial like that are the best ones - they know exactly what they want, and we don't have to guess.
The price comparison people get wrong
Almost everyone compares a €20 subscription with the cost of automation and reaches a verdict in two seconds. But that's comparing apples and oranges.
The subscription is the price of a tool. A person still does the work, and their hour costs the business about $46.89. With automation you're not buying a tool, you're buying work done - in our model, up to $10 an hour, maintenance included.
So the right question isn't "which is cheaper per month" but "how many hours a month does this job take". At a few hours, the subscription is cheaper. At a few dozen, the agent is. More on how pricing is built here, and on how an agent differs from a chatbot here.
In short
- ChatGPT - when the work is occasional, varied and done by a person.
- An AI agent - when the work repeats, follows rules and has to happen without a person.
- Both - the most common real answer. The team uses ChatGPT for everyday thinking, and agents handle what repeats.
How to tell which work is worth handing to an agent is covered in a separate article.
Questions about ChatGPT and AI agents
Can I just use ChatGPT instead of an AI agent?
It depends on who presses the button. ChatGPT is a tool a person starts every time: they open the window, enter the data, read the answer and copy it to where it needs to go. An AI agent is a system that runs the same chain on its own, without a person, connected to your email, CRM or accounting. If you do the job a few times a week and don't mind copying by hand, ChatGPT is perfectly enough and you don't need to pay for automation.
How is an AI agent different from ChatGPT?
In three ways. First, initiative: ChatGPT waits for you to type, while an agent reacts to an event on its own - an incoming email, a new order, an approaching deadline. Second, access: ChatGPT can't see your email, CRM or inventory, while an agent is connected to them. Third, the result: ChatGPT's answer stays in the chat window, while an agent's result lands in the system or with the customer.
When is ChatGPT no longer enough?
When the task repeats and a person has to be involved every single time. In practice the line shows up at a few dozen identical cases a week: ChatGPT shortens the thinking part, but copying, checking and entering stay with the person, so only part of the time is saved. When the same work adds up to dozens of hours a month, that remaining part becomes the most expensive one.
Can't I connect ChatGPT to my systems and files?
Partly, yes - there are file uploads, projects, custom GPTs and connectors to some services. But four differences remain: someone still has to start the chat, there's no guarantee a hundred cases will be handled the same way, there's no action log you could hand to an auditor, and there's no error handling when data is missing or a service doesn't respond. In automation, those boring things are exactly the point.
Which is cheaper: a ChatGPT subscription or an AI agent?
A ChatGPT subscription costs roughly €20 to €25 a month per person, so on the invoice it will always be smaller. But you should compare the cost of the whole job, not the subscriptions: with ChatGPT a person stays in the loop, and their hour costs a US business about $46.89. In our model an hour of AI agent work costs up to $10, with a $300 monthly minimum. So the answer depends on how many hours a month the work adds up to: at a few hours the subscription is cheaper, at a few dozen the agent is.
Where should I start if I'm not sure which one I need?
Start with ChatGPT and do the job by hand for a few weeks. It's the cheapest way to find out whether the work can be automated at all: if ChatGPT handles it when you give it the right data, an agent will handle it automatically. If ChatGPT can't, an agent can't either - it'll just cost more. You also come out of it with a technical brief and a clear number of hours.