AI ResistanceA Practical Guide to Managing Resistance to AI in Organizations

01The essentials

• There isn’t one kind of resistance to AI. There are at least nine. Each one is afraid of losing something different: a job, data, budget, control, the quality of the work…
• The three things you hear most often don’t mean the same thing. “AI kills critical thinking” is a concern backed by research. “AI is going to take my job” is fear. “That’s for low-level work” is the opposite of fear: it’s playing AI down.
• The map sorts ways of arguing, not people. The same person can take different stances depending on the topic or the day.
• Almost every stance has a point. If you treat them as excuses, the resistance doesn’t go away: it goes underground.
• Inside a company, the most accepted way of saying no is caution. “Not yet”, “not without safeguards”, “show me the return”.
• Alliances shift depending on what AI is used for. People worried about data and people worried about their jobs have nothing in common… until AI programs start working on their own with client data.
• Counting how many people use AI isn’t enough to know whether change is working. Look at whether objections leave the corridor and get discussed openly.

02Why this guide

In fall 2026, scenarioDNA, a cultural analysis firm based in New York, published The Subcultures of AI Doomerism. It is a map of six groups that, according to its analysis, are shaping the public debate about artificial intelligence.

The report’s main idea is very useful: people who distrust AI use the same words (“control”, “safety”, “access”), but they aren’t talking about the same thing, asking for the same thing or proposing the same thing. These are the six groups, with the names scenarioDNA uses:

• Thresholdists: want to put limits on the most advanced AI capabilities until we are sure we can control them.
• Human Preservationists: defend human work, authorship and people’s ability to decide for themselves.
• Material Refusers: oppose the physical cost of AI: the water, energy and land that data centers consume.
• Strategic Containmentists: see AI as a matter of national security and control over chips and models.
• Commons Technologists: argue for open AI that anyone can inspect, modify and run on their own machines.
• Normalizers: aren’t pessimists at all; they propose bringing AI into organizations with rules and good practice.

Throughout this guide we keep these names in italics so they don’t get confused with the nine stances we found inside organizations.

A company doesn’t work like society

scenarioDNA’s map describes the public debate. Inside a company the questions are much closer to home: do I use the tool on Monday, who decides what gets bought, what happens to my job, who signs the report that AI helped write. Three things change when you move from society to a company:

• There are power relationships. Hierarchies, contracts, clients, budgets and performance reviews. Saying what you think has consequences.
• People don’t say the same thing everywhere. A concern can show up in meeting minutes as “regulatory risk” and in the team’s group chat as “they want to replace us”.
• Getting it wrong has an immediate cost. Nobody is debating the future of humanity; they’re debating the quarter close, a client or this week’s workload.

Neither “doom” nor “excuses”

This guide started with three sentences you hear in almost any organization: “AI kills critical thinking”, “AI is going to take my job” and “AI does low-level work; we’re a boutique firm”. It’s tempting to call these “doom-laden attitudes” or “excuses”. Both labels are unfair and inaccurate.

It isn’t doom, because only one of the three sentences expresses fear. The first is a concern grounded in research. The third is the opposite of doom: it doesn’t fear AI, it looks down on it. And they aren’t excuses, because almost all of them have a point. Calling them excuses also backfires: people who feel dismissed don’t stop resisting, they just stop saying it out loud.

That’s why we talk about subcultures of resistance: recognizable ways, shared by groups of people, of saying “no”, “not yet” or “not like this”.

03How the map was built

This section explains where the map comes from and how it was put together. If you’d rather go straight to the nine stances, skip to section 4.

Where the stances come from

The map is based on fieldwork we carried out between July and October 2026. It combines two sources:

• Interviews with people in different roles about how the arrival of AI is being experienced in their organization.
• Situations from consulting and training projects: what gets said in meetings, workshops and working sessions when AI enters day-to-day work.

Most of the organizations were small and medium-sized businesses. Larger organizations and even a multinational company also took part. That variety let us see which stances repeat across different settings, but it doesn’t turn the map into a general rule: in each organization, the weight of each stance will be different.

The research was led by Alfonso Romay and Cinta Vergés of 1North.

What a culture map is

A culture map is a way of organizing what people say in order to understand what lies behind it. It doesn’t measure opinions with a survey: it analyzes language. It looks at the words, stock phrases and images that keep coming up in a group, because they reveal how that group thinks and what it cares about.

To organize what we saw and heard we drew on scenarioDNA’s culture mapping, which in turn builds on Raymond Williams’s categories (1977). It consists of placing each way of talking on a two-axis grid and watching how it moves over time. scenarioDNA holds a United States patent on a computer system that analyses the language of internet accounts. This guide does not use that system: the analysis is ours, built from interviews and observation.

One rule of the method is especially useful inside a company: group by type of argument, not by how often you hear it. What gets repeated most isn’t always what carries most weight. The opinion heard most often in leadership meetings is rarely the one with most influence on day-to-day work.

The two axes

Vertical axis: official culture ↔ subculture. It shows where something is said. At the top is official culture: what gets said in leadership meetings, in documents and in announcements. At the bottom is subculture: what gets said in the corridor, over coffee or in the team’s group chat. It doesn’t depend on rank: a director can talk in the corridor and a junior can bring an issue to a committee.

Example: the same worry about jobs shows up in official culture as “we need to manage the impact on staff” and in subculture as “they want to replace us”.

Horizontal axis: analytical codes ↔ expressive codes. It shows what kind of argument is used to defend a stance. The method uses the word “codes” for the type of language people use. Analytical codes rely on rules, data, money or errors. Expressive codes rely on identity, fear, pride or fatigue.

Example: “GDPR doesn’t allow it” is an analytical argument. “We’re not just another company” is an expressive one.

The four quadrants

Crossing the two axes gives four quadrants. Their names come from the British cultural critic Raymond Williams. In 1977, Williams distinguished three kinds of elements in any culture:

• dominant: what organizes the present;
• residual: what was inherited from the past and is still active;
• emergent: what is new and starting to appear.

scenarioDNA adds a fourth: disruptive, which challenges what is established. Applied to AI inside a company, the four quadrants mean this:

Stances, not people

Each point on the map is a stance: a recognizable way of arguing about AI, with its own concerns, words and logic. scenarioDNA calls them “formations”. They are not types of people, and that has two practical consequences:

• The map isn’t for labeling anyone. Doing so creates pushback and is also inaccurate, because people change stance depending on the topic and the moment.
• The goal isn’t to “win over the resisters”, but to respond to each argument on its own terms.

How we tell one stance from another

For the map to be useful, the stances shouldn’t overlap. To check, we ask each one two questions:

• What is it afraid of losing? A job, the company’s reputation, data, budget, decision-making power, the ability to think… The method calls this the protected object.
• What is its logic in one sentence? A sentence along the lines of “because this happens, I do that”. The method calls this the core grammar.

Example: Custodians and Verifiers both talk about “control”. But Custodians fear a legal or data problem, while Verifiers fear handing in work with mistakes. Because they’re afraid of losing different things, they are different stances.

What each profile includes

We describe the nine stances using the same headings. The first six follow scenarioDNA’s model. We added the last two, because a guide for companies isn’t only for understanding but for acting.

1. The sentence that sums it up.
2. What you hear: literal phrases that give the stance away.
3. What they’re afraid of losing.
4. Words and images they use.
5. Who they lean on and who they clash with.
6. Their logic in one sentence.
7. Where they have a point.
8. What works with them.

04The map

This is the map of the nine stances. Each point is a way of arguing against AI (or of delaying it). Its position shows where it is said and with what kind of argument.

How to read the map

• On the left are the kinds of resistance that sound reasonable: rules, figures, errors. They’re easy to defend in public and, for that very reason, hard to argue against.
• On the right are the ones about identity and emotions. They’re harder to put into words, but they have more influence on what people actually do.
• The bottom half is what gets said in the corridor. It’s what leadership usually doesn’t see until it has already become a problem.

05The nine stances

Here are the nine stances, ordered by quadrant. First, a one-line summary of each:

• The CustodiansSlow things down because of legal and data protection risk.
• The Wait-and-SeersWait for others to prove that AI pays off.
• The ArtisansBelieve AI isn’t up to the standard of their work.
• The TerritorialsWant to control who makes decisions about AI.
• The Window-DressersAnnounce AI, but don’t change anything.
• The VerifiersDon’t trust what AI produces.
• The Guardians of JudgmentWorry that people will stop thinking for themselves.
• The ThreatenedAre afraid of losing their jobs.
• The OverloadedHave no time or energy for yet another change.

Each profile covers how the stance works, where it has a point, what happens if it’s ignored, what works and what doesn’t, and a few questions to spot it in your organization.

06The stances side by side

The profiles describe each stance on its own. Looking at them together reveals things you can’t see one at a time.

What each one is afraid of losing

The nine stances fall into three levels, depending on what they’re afraid of losing. This is what sets them apart: behind the same question (“do we use AI or not?”) there are very different concerns.

Same word, different meaning

When leadership talks about “quality”, “risk” or “control”, each stance hears something different. The same message is heard nine ways. That’s why, when explaining change, it works better to organize messages around each stance’s concern than around departments.

Alliances shift depending on what AI is used for

The stances don’t have fixed allies. Depending on what AI is doing in the organization, some objections are heard more than others, and alliances nobody expected appear.

What changes compared with scenarioDNA’s map

Comparing this map with scenarioDNA’s shows what happens to the big stances in society when they enter a company.

The most striking change is in the Human Preservationists. On scenarioDNA’s map they are a single group. Inside a company they split into three, depending on where people are speaking from:

• At the base, fear of losing one’s job: the Threatened, in the Emergent quadrant.
• In the middle, concern about learning: the Guardians of Judgment, in the Disruptive quadrant.
• In leadership, defense of the brand: the Artisans, in the Dominant quadrant.

One underlying concern, the defense of human work, is expressed in almost opposite languages. That’s why a single communication campaign for everyone rarely works: it only reaches one of the three groups.

The type of company culture matters too

The stances don’t show up the same way in every organization. To relate them to each company’s type of culture we use Kim Cameron and Robert Quinn’s Competing Values Framework. This model sorts cultures into four types based on two questions: does the organization prefer flexibility or control, and does it look more inward (people, processes) or outward (clients, market)? Its questionnaire, the OCAI, measures both the current culture and the culture the company would like to have.

07What to expect over the next three years

scenarioDNA’s map looks thirty years ahead. Inside a company, it’s more useful to think two or three years ahead. We propose three stages. In each one, some objections are heard more than others because AI touches different things. It’s a hypothesis: each organization will move at its own pace.

How to know if change is working

Lots of people using AI isn’t enough. The best sign that change is going well is that the stances move on the map over time. The method calls this movement migration. Here are four signs that change is working:

• Objections are raised in meetings, not just in the corridor. What is said openly can be dealt with.
• The Custodians move from “we can’t” to “this is how we can”. They stop blocking and write the usage policy.
• The Verifiers go from skeptics to go-to people. They show others which tasks make checking the result quick.
• The Window-Dressers stop counting licenses and start counting real use.

Some concrete indicators to track it:

• How long it takes to approve a new use of AI.
• What percentage of licenses is used every week.
• How many processes now work differently.
• What proportion of people talk openly about how they use AI.
• How many objections arrive through official channels and how many only through the corridor.

08Self-check: which stances are around you?

Tick the phrases you’ve heard in your organization over the past month. It isn’t a full diagnosis, but it will give you a first idea of which stances are most present. Nothing you tick leaves your browser: it isn’t saved or sent anywhere.

09How to apply the map in your organization

The map in this guide reflects what we’ve seen in a range of organizations. To find out what it looks like in yours (which stances carry most weight and which barely appear) we propose six steps, adapted from scenarioDNA’s method.

1. Collect real phrases

   Gather literal phrases about AI, exactly as people say them, not summaries of opinions. You can get them from interviews, discussion groups, a sentence-completion survey or leadership announcements. Internal channels (chats, forums) should only be used with consent and anonymously. For each phrase, note who says it (the role, never the name) and where (in a meeting, over coffee, in a document).

   Some sentence starters that work well: “AI in my job is…”, “What worries me most about using AI is…”, “Here, decisions about AI are made by…”, “When someone on my team uses AI…”.

   And some interview questions: “What have you heard people say about AI in the last few weeks, and where?”, “What would your team lose if AI worked perfectly?”, “Who should decide how it’s used?”.

2. Place each phrase on the map

   Score each phrase from −5 to +5 on the two axes. You can use this as a guide:

   • Vertical axis (where it’s said): +5 if it appears in an official document or a committee; +2 if it’s said in a team meeting with the manager present; −2 if it’s said between colleagues; −5 if it’s only said in confidence.
   • Horizontal axis (what kind of argument): −5 if it cites a rule, a piece of data or a figure; −2 if it’s a technical or first-hand argument; +2 if it appeals to values or identity; +5 if it expresses a clear emotion, such as fear, pride or fatigue.

   Ideally, two people score separately and then compare and discuss the differences.

3. Group by stance, not by how often you hear it

   Group the phrases by what they’re afraid of losing and by their logic. A single well-argued phrase can reveal a new stance. And a hundred people repeating the same slogan are still just one stance.

4. Write a profile for each stance and test it

   Use the eight headings from section 3 and review the profiles in a workshop with people from the organization. If someone doesn’t recognize themselves in the profile of “their” stance, the profile is wrong.

5. Cross the stances

   Look at what each one is afraid of losing, which words they share with different meanings, which alliances appear at each stage and how they fit with the company’s type of culture (measured with the OCAI).

6. Repeat the exercise after 6–12 months

   Run the same process again with the same method. Comparing where each stance was and where it is now (what the method calls migration) is the most valuable result of the whole exercise.

Ethics and privacy

• Real anonymity: no phrase can be linked to a specific person.
• Explain what the information is for and ask for consent. Collect only what you need and comply with the GDPR.
• The map must never be used to assess anyone’s performance or to make decisions about specific people.

Common mistakes

• Labeling people (“John is one of the Overloaded”) instead of analyzing arguments.
• Using the map to look for “enemies” of change.
• Treating objections as excuses and ignoring the part where they have a point.
• Counting how often something is repeated instead of identifying stances, and taking the loudest voice at face value.
• Doing it only with leadership. The map would be skewed toward what gets said in meetings.
• Stopping at the map without deciding what to do about each stance.

10What this map doesn’t say

• These are hypotheses, not rules. The stances come from interviews and from situations in consulting and training projects between July and October 2026. They describe what we observed, not what happens in every organization. The positions on the map sum up that material and need to be tested case by case.
• The balance changes from one organization to another. In one company the Custodians may carry a lot of weight and the Artisans almost none, and in another it may be the other way round. The map tells you which stances to look for, not how much each one weighs in your organization.
• It has a context. Most of the organizations in the fieldwork were small and medium-sized businesses, although larger organizations and a multinational also took part. In settings that were less represented, other stances may appear.
• There are stances we’ve left out. We did see them, but they were uncommon in the fieldwork: the Sovereigntists (“no big-tech tools, let’s build our own”), the equivalent of the Commons Technologists, and the Objectors (“it isn’t ethical”, “it uses too much water”, “it steals from creators”).
• It isn’t a personality test or a tool for assessing people.
• The figures we cite are indications, not absolute truths. They come from studies with known limitations: surveys where each person rates their own behavior, samples that aren’t representative or debatable ways of measuring return.

11Glossary

Stance

A recognizable way of arguing about AI, with its own concerns, words and logic. It isn’t a type of person. scenarioDNA calls them “formations”.

Culture map (culture mapping)

A method for placing what people say on a two-axis grid and watching how it changes over time.

Official culture

What gets said through formal channels: leadership meetings, documents, announcements. The top half of the map.

Subculture

What gets said through informal channels: the corridor, over coffee, the team chat. The bottom half of the map.

Analytical codes

Arguments based on rules, data, money or errors. The left side of the map.

Expressive codes

Arguments based on identity, fear, pride or fatigue. The right side of the map.

Residual

The quadrant of official culture and analytical codes. What was inherited and is still active: slowing things down by leaning on rules and proven cases.

Dominant

The quadrant of official culture and expressive codes. What organizes the present: defending the company’s identity and who decides.

Disruptive

The quadrant of subculture and analytical codes. What challenges the established order with technical or first-hand arguments.

Emergent

The quadrant of subculture and expressive codes. What is new and starting to appear: fears and fatigue that will end up as requests.

Protected object

What a stance is afraid of losing. Used to tell stances apart.

Core grammar

A stance’s logic summed up in one sentence, along the lines of “because this happens, I do that”. Also used to tell stances apart.

Migration

The movement of a stance on the map over time. The best way to know whether change is working.

Shadow AI

Using AI tools that the company hasn’t authorized to do work tasks.

AI agent

An AI program that carries out tasks on its own, without a person stepping in at each step.

Hallucination

An AI error where it invents data or sources and presents them as true.

OCAI

Cameron and Quinn’s questionnaire (Organizational Culture Assessment Instrument), which measures an organization’s current culture and the culture it would like to have.

12References

1. scenarioDNA (2026). The Subcultures of AI Doomerism. StoryEngine · Cultural Intelligence by scenarioDNA, fall 2026. On the method: scenariodna.com/culturemapping.
2. Stock, T. J. and Stock, M. L. (2015). System and method for culture mapping. U.S. Patent 9,002,755. Google Patents.
3. Williams, R. (1977). Marxism and Literature. Oxford University Press.
4. Cameron, K. S. and Quinn, R. E. (2011). Diagnosing and Changing Organizational Culture: Based on the Competing Values Framework (3rd ed.). Jossey-Bass.
5. Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R. and Wilson, N. (2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. CHI 2025. Microsoft Research.
6. Challapally, A., Pease, C., Raskar, R. and Chari, P. (2025). The GenAI Divide: State of AI in Business 2025. MIT NANDA. PDF.
7. Microsoft and LinkedIn (2024). 2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part. Microsoft.
8. O Morain, C. and Aykens, P. (2023). “Employees Are Losing Patience with Change Initiatives”. Harvard Business Review. hbr.org.