What AI is, what automation is, and why calling everything AI is noise

At some point in the last three years, the term “AI” started sticking to almost anything: an electric toothbrush, an invoicing program that has existed since 2015, a vending machine, a soft drink. When a word describes everything, it stops describing anything.

For a business trying to understand what it is buying, that noise has a concrete cost. You pay artificial-intelligence prices for a familiar rule, or dismiss a tool that would have helped because it sounds just like the other twenty. Distinguishing the two is not a vocabulary debate: it determines whether a project makes sense before you spend a euro.

Automation is not the same as using AI

The difference fits in one sentence: rule-based automation executes instructions written by a person, while an AI system interprets data and makes decisions based on probabilities, with a margin of error. The rest of this article unpacks that sentence.

Automating means a machine carries out a task a person used to do, following instructions. Those instructions can be of two very different kinds, and the difference matters much more than it seems.

Rule-based automation works with “if this happens, do that” logic, explicitly written by a person. A bank that flags every overseas transfer above €5,000 as suspicious is using a rule, not artificial intelligence. These systems are transparent, easy to audit and perfectly suited to anyone who needs complete traceability: accounting, payroll, inventory control. Their limit is that they can only do exactly what they were programmed to do, and nothing more.

Artificial intelligence does not follow a fixed recipe: it identifies patterns in data and makes decisions based on probabilities. It can read an unstructured customer email, understand its tone and decide how to prioritise it, something a rule-based system cannot do because no rule can anticipate every possible way of writing a complaint. That is its value, and also its condition: where there is probability, there is error, so someone needs to review the result.

Neither approach is better in the abstract. A rule wins when tolerance for error is minimal and every decision needs a step-by-step explanation. AI wins when data is varied, changing or does not fit a fixed format. Most well-designed systems combine the two: rules where control is needed, AI where adaptation is needed.

Three small-business examples: what is a rule and what is AI

In the abstract, the distinction sounds academic. Three tasks from a small business make it clear in a minute.

• Recurring invoices. Forty customers with the same monthly fee, invoiced on the first, sent by email and entered in the accounts. It is a rule from beginning to end: there is nothing to interpret and no AI is needed. If something is sold as “smart invoicing” to solve this, you are buying ordinary invoicing software with a new label on it.
• Reading a scanned invoice from a new supplier. Each supplier lays it out differently, and you need to extract the net amount, VAT and description. That is AI: no hand-written rule anticipates every possible format. And because it works with probabilities, a sound design includes a person reviewing low-confidence results, not “the system will take care of it”.
• Routing email from a shared address. Distinguishing an urgent complaint from a quotation request or an attached invoice involves interpretation, so that classification is AI. What follows — who gets each item, at what priority, and how quickly they must reply — is a rule again. The good answer to “is this AI or automation?” is almost always “both, in that order”.

The useful thing about looking at it this way is that the question stops being which technology to buy and becomes where interpretation is needed. In the parts where nothing needs interpreting, AI adds cost, opacity and a probability of error that did not previously exist.

AI washing in practice

This confusion of terms has a deliberate version, and it has a name: AI washing. It means promoting a product or service by exaggerating the role artificial intelligence actually plays in it. You do not have to lie outright: calling a conditional form an “intelligent agent”, or calling the same sorting algorithm the product used eight years ago an “AI engine”, is enough.

The phenomenon extends beyond product marketing. A Resume.org survey of 1,000 hiring managers in the United States, published in April 2026, gives an inside view. Among those planning layoffs, AI is the most frequently cited reason: 44%, ahead of internal reorganisation (42%) and budget constraints (39%).

The actual effect they report is different. Only 9% say AI has completely replaced any role; around 45% say it has partly reduced the need to hire, and another 45% report little or no effect on workforce size.

The third figure closes the circle: 59% acknowledge that, when explaining a hiring freeze or a layoff, they emphasise AI because it sounds better to their audiences than financial constraints. It is AI washing measured from within, which is why that number matters: it does not measure what AI does in businesses, but what naming it achieves.

Peter Cohan, associate professor of management practice at Babson College, summed it up in Built In with a phrase that is hard to improve on: blaming AI is “the least bad reason” a company can use, because it makes it look more innovative and forward-looking than admitting a financial problem.

It is worth saying what those numbers are and are not. They are US data from a labour market unlike Spain’s, and there is no equivalent survey here we can cite; nor are they our data, as we do not publish customer results.

What travels well between countries is not the percentage: it is the incentive. Saying “AI” improves the story, and while it does, some people will say it without any AI behind it.

Why this creates noise, not just confusion

When anything is called AI — a spreadsheet with a formula, a sorting algorithm from 2015, a conversational assistant trained with billions of parameters — the word stops being information and becomes decoration.

And noise is not just irritating: it costs money in both directions. People buy AI where a rule would have been enough, and reject a simple automation that would have solved the month’s problem because it came wrapped in the same pitch that had already disappointed them twice.

There is a third cost, less visible and more troublesome in the medium term. If nobody knows which part of the system decides on its own, nobody knows what needs watching, who is responsible when it gets something wrong, or what to tell the customer. Those three things are not clarified by reading a brochure: they are clarified by asking.

The way out, then, is a question, and it is not “does this have AI?”. It is “what kind of decision does this system make, with what data, and what happens when it gets it wrong?”. It works equally well for a three-line rule and the most sophisticated model on the market, and separates substance from packaging without requiring any technical knowledge. It is also the first of the questions worth asking a supplier before signing, where you will find the other four.

Most processes businesses bring us contain both halves: one stretch of pure rules and another where something needs interpreting. Knowing which is which is not a technical detail; it is what determines the budget. And you find out by looking at the process, not the product name.