Four months ago we started writing an internal note. It was not a report and it was not a product: it was a jotting for ourselves, to put in writing something that happened in almost every project. You turn up with a redesigned process, with artificial intelligence built into it, and the process is not the problem. The problem is that somebody in the room says “not yet” and you do not quite know what sits behind that “not yet”.
That note has ended up as Resistenc-IA, a public guide of about forty-five minutes’ reading, under an open licence and with two PDFs to download. This is how it happened.
Four months listening
The fieldwork ran from July to October 2026 and went through more than fifteen companies: mostly Spanish small and mid-sized firms of very different sizes, plus one multinational. We did not run a survey. We kept noting down how people say no to artificial intelligence inside an organization: in meetings, in committees, in corridors and in emails.
What repeated itself was not the arguments but the stances. Ten of them, in the end. The same sentence, “this is not mature yet”, means different things depending on who says it and where they say it from. And each one calls for a different answer.
The piece we were missing
We had plenty of observations and no way to order them. We found it reading scenarioDNA, a cultural analysis firm in New York founded by Tim Stock and Marie Lena Tupot. In their publication Analyzing Trends they put out The Subcultures of AI Doomerism: a map of six groups that, by their analysis, are shaping the public debate about artificial intelligence.
Their central idea solved our problem. People who distrust AI use the same words —“control”, “safety”, “access”— but they are not talking about the same thing, nor asking for the same thing, nor proposing the same thing. That was exactly the tangle in front of us inside companies, only at corridor scale instead of public-debate scale.
The same word, said by two different people, can be asking for two opposite things. That is where half of these projects get stuck.
Their method is called culture mapping and it even has a registered patent. We have not applied it: it inspired us to build our own map, with two axes and four quadrants, over what we had heard. We have no commercial relationship with them, and that is worth saying.
We sent it to the people who started it
With the draft put together we did what seemed the decent thing: we sent it to the scenarioDNA team so they could see what we had done with their idea. They replied, reviewed the methodology and gave us their approval. The thanks go to Marie Lena Tupot, who took the trouble to read it closely.
The authors of the original method approving how you used it is not the same as saying the work is theirs. The research, the stances and the conclusions are ours, and so are the mistakes. But knowing the framework is properly applied, you publish with a steadier hand.
Ten in the end
For most of the work we had nine stances. The tenth turned up late and is the most uncomfortable of them: the rule-benders. They do not reject artificial intelligence, they reject the way their company has rolled it out. They use their own tools, skip approvals and adapt the rules on their own terms.
What is interesting is how they justify it, and it has had a name for centuries: epikeia, reading a rule by its purpose rather than its letter. “The rule is there to protect data; there is no sensitive data here, so I will go ahead.” The question it leaves open is when the exception stops being an exception and becomes the normal way of working. That is why it sits in the emergent quadrant: it is not breaking a rule once, it is a way of working shared below the official system that ends up creating rules of its own.
It is also the stance closest to what IT departments call shadow IT, and the one that fits worst with the idea that resistance to AI means slowing it down. Here the resistance moves fast; it just moves outside the lane.
Why we published it openly
We could have kept the guide inside and used it ourselves. The opposite seemed the better idea. There are a lot of people right now in transformation projects with AI in the middle of them —training, innovation, general management, strategy, systems— wrestling with the same thing. A map is not used up by sharing it.
So it goes out under a Creative Commons BY-NC-SA 4.0 licence: you can use it, adapt it and take it into your organization, crediting where it came from and without making a business of it. There is also a poster to print and a set of slides to take it into a working session.
What is missing
The guide looks at only half the board. Organizations also say “yes” to artificial intelligence, and they do not say it in one single way or for one single reason either. We are preparing Impuls-IA, the companion guide to the stances in favour, and a conclusions document with whatever comes out of putting both halves together.
In the meantime, if you are in the middle of one of these projects and you recognise any of the ten stances in your own place, tell us about it. A map like this gets sharper with real cases, not in an office.