Automation or AI: what is the difference?

The short answer

Automation or AI: what is the difference?

The difference between automation and AI lies in who comes up with the rule. With automation, a person writes out the steps: if the invoice is above a certain amount, it goes to the director. With AI, you give a goal and examples, and the system works out for itself what to do with new input.

For you as an owner, that makes a real difference: a fixed rule does exactly the same tomorrow as it did today. AI can read an email worded differently from every one before it and still do something sensible with it. The other side is that with AI, you have to keep checking what comes out.

The question comes up more often now that more businesses use AI. According to preliminary figures from Statistics Netherlands (CBS), one in six Dutch businesses with two or more people working there used AI in 2025, up from 13 percent a year earlier.

The definition

Where does the law draw the line between software and AI?

The European AI Act draws the line at the ability to infer: software that only carries out rules set by people falls outside it, according to the legislator. Recital 12 says that the capability to infer is a key characteristic of AI systems.

The same recital says that systems based on rules defined solely by natural persons are not AI. The European Commission worked this out in guidelines in July 2025. As examples of software outside the definition, it names a database that sorts or filters data, standard spreadsheet software without AI functions, and a prediction that simply takes last week’s average.

That line is useful even if you never need the law itself. The question is always: did someone write the rule down, or does the system find it itself? How clever something looks says little about that.

Where rules suffice

When is ordinary automation enough?

Ordinary automation is enough when you can capture the step fully in rules and the input always looks the same. A fixed rule is then cheaper to maintain, predictable and easy to test.

Take the monthly report: collecting figures from a handful of systems and putting them neatly into a fixed format. That is many steps, but they follow rules that are fixed.

Copying an address from one list to another is another such step. AI adds little there and does bring in uncertainty.

My reading

Why are simple steps sometimes not simple?

Simple steps are sometimes not simple because they hold years of experience that nobody ever wrote down. I call that hidden craftsmanship, and that is exactly where ordinary automation stops and the question of whether AI can help begins.

Part of a quote is rule work: looking up materials, adding up hours, putting the text in the house style. But in between are steps that look the same and are not. In my book, I put it like this:

Judging what this customer needs, sensing where the risk lies in this particular job, choosing which margin fits this particular relationship. It is one line in the same system, and there are years of experience in it.Paul Hoffman

An if-then rule cannot capture that, because the craftsman often cannot explain it in rules himself. AI can do something with it, provided you first set down what good work looks like here. That is why AI, to me, is a redesign of the work, not a purchase.

What goes wrong

What goes wrong if you choose the wrong one?

If you choose the wrong one, it goes wrong in two ways: fixed rules break on every exception, and AI sometimes gives a wrong answer with great confidence. Both can be avoided if you choose per step.

Rules that you extend for every exception grow into a tangle nobody dares to touch anymore. That is usually the moment someone says AI should be added.

Kalai and three colleagues, researchers at OpenAI and Georgia Tech, showed in September 2025 that language models guess because training and evaluation reward a guess over admitting they do not know. If you put AI on a step where a fixed rule suffices, you take on that risk for nothing. If you put AI on a step that needs judgment, have a person sign off on what goes out the door.

In practice

How do you choose per step?

Choose per step by writing out one workflow and asking at each step: is the rule fixed, or does someone have to make a judgment? What is fixed, you automate; where judgment comes in, you look at whether AI can help.

Set aside an hour for it with one employee who knows the work. Put the steps on a whiteboard, from request to sent, and mark each step with an R for rule or a J for judgment. A step with a J is where the craftsmanship sits, and so where a person keeps signing off at the end.

Want to know how far your business has come? Read about the five levels of AI in a business and take the short check there.