Resistance to AI in the workplace, and what to do about it

Resistance to AI in the workplace is rarely about the technology. It is about uncertainty: will I keep my job, will a system start judging me, does my craft still count, and who does the extra work while it is being introduced. Much of that worry is reasonable. You address it by saying early what you do and do not plan to use AI for, by letting the people who do the work help choose where to start, by keeping the final say with a person and by measuring whether it is really used. If you have a works council, it belongs at the table early, and sometimes it has a right of consent.

This piece is written for the managing director or owner of a company of 50 to 350 people who wants to introduce AI and notices that it is not going smoothly. Or who expects that and wants to prevent it.

A few related topics are covered elsewhere, and we do not repeat them here. How to have the first conversation about jobs is in Getting more from the people you already have. What a tool costs that nobody uses is in The hidden costs of an AI tool nobody uses. Who needs to learn what is in Who in your organisation needs to learn AI, and who does not. This piece covers what sits in between: how to recognise resistance, where it comes from and what to do about it.

What the resistance is really about

When people respond to AI with reluctance, it is tempting to think they do not understand the technology or do not want to learn. That is usually not the problem. If you listen carefully, you hear five other concerns.

My job. Most news about AI is about jobs disappearing. So it makes sense that someone who does a recurring task wonders whether that task is their job. As long as management says nothing about it, everyone fills in the answer for themselves.

Being judged. A system that prepares work also sees how fast someone works, how many mistakes they make and when they log in. People sense that. Someone who asks what happens to that data is asking a question you would ask in their place.

Craft and status. The planner who has learned over twenty years which customer always calls late on a Friday is somebody in the company because they know that. When a system takes over part of that work, it feels like a loss of value. Even when nobody meant it that way.

Extra work during the rollout. A new way of working costs time first. The old work carries on, testing and explaining are added, and the promised time savings come later. Anyone who has been through that a few times does the sums in advance.

Earlier IT projects. Almost every company has had a system that was going to make everything better and did not. A move to a new ERP package that ran a year over, a portal nobody logs into. People remember that. They wait to see whether this time is different, and that is sensible.

None of these five is about whether AI works. They are about what it means for a person. That is good news for a director, because you can answer those questions. A vendor cannot.

Two colleagues talking at their desks. Most concerns about AI come out in a one-to-one conversation, rarely in a full meeting.

Three forms of resistance, and how to recognise them

Resistance does not always look like resistance. There are three forms, and the hardest one is the quietest.

Open resistance feels the most uncomfortable, and it is the easiest to work with. Someone who says out loud what bothers them is giving you information. Often the critic is also the person who knows the process best. Bring that person into the choice and they often become the best tester you have.

Silent resistance only shows when you look for it. A meeting where everyone agrees and nobody asks anything says little about support. In a group, almost nobody is the first to say they are worried. A conversation with two or three people separately yields more than a kick-off with the whole team.

Passive non-use is the most expensive form, because it stays invisible for a long time. On paper the tool has been introduced. In practice people do the work the old way, and the data in the new system stops being right. What that costs is in the piece on hidden costs. Here the question is the cause. It almost always lies with a tool that does not fit how the work really goes, and with people who were never asked how the work really goes.

Why much of the resistance is reasonable

It is tempting to see resistance as something to overcome. First check whether it is right.

Employees are not against AI as such. In a Eurobarometer survey by the European Commission, published in February 2025, more than 60 percent of Europeans view robots and AI at work positively. Over 70 percent believe they improve productivity. At the same time, 84 percent think AI in the workplace requires careful management to protect privacy and ensure transparency. That is the tone you also hear on the shop floor: fine, if it is properly arranged.

Companies themselves hesitate for reasonable reasons too. Statistics Netherlands (CBS) asked companies that had considered AI but did not use it why they did not. Lack of experience was mentioned most often, by 73 percent. Privacy was mentioned by 49 percent, uncertainty about the legal consequences by 42 percent. If a management team may have those concerns, an employee may have them too.

And anyone who has seen an IT project fail before has a point. Many rollouts get stuck on the same things: the people who do the work were not at the table, no time was set aside, and nobody checked after a month whether it was being used. Someone who waits for that reason is showing that they pay attention.

That does not mean every concern is justified. Sometimes a fear is based on a news story that has nothing to do with your company. You only know once you ask. The rule of thumb: treat every concern as information first. It often tells you something about your plan that you had not seen yourself.

Seven steps that help

None of these steps is complicated. They mainly ask for management attention, which is the scarcest resource in a company of this size. Do them anyway. They cost less than a tool nobody uses.

1. Say early what it is and what it is not. Do not wait until a tool has been chosen. Say at the start what you want to achieve, and also what you do not plan to do. For example: we want to retype fewer orders, and we will not use the system to assess individual performance. Whatever you leave unsaid, people fill in themselves, usually with the worst case.

2. Let the people who do the work help choose. Whoever runs a process every day knows where time leaks away, where mistakes arise and which exceptions a system needs to know about. Let those people have a say in which application comes first. That leads to a better choice. And whoever helped choose has a stake in making it work.

3. Say out loud that nobody is being replaced, and make it concrete. The aim is to take work off someone's plate, not to replace someone. Say that, and also say what happens with the time that frees up. More customer contact, clearing the backlog, the work that is now left undone. A promise without substance is not believed. Only promise what you can deliver. If you are not sure about something yet, say that too.

4. Start where people feel the pain. Do not ask what AI can do for your team. Ask which part of the work they would most like to be rid of. Almost everyone has an answer: the retyping, the looking things up, the same email again and again. A first application that solves a chore like that convinces people by itself. A first application that mainly helps management makes the second project harder.

5. Keep the final say with a person. Agree that the system prepares and a person decides, or that a person watches and can always step in. That second option is called human on the loop: the system runs on its own, a person supervises and can overrule it. It removes much of the worry about craft, because the judgement stays with the craftsperson. It also keeps the risk manageable. Record per type of decision who has that final say, in your AI policy.

6. Make time for the rollout. The extra work during a rollout is real. Acknowledge it and make room for it. Take something else off the schedule in the weeks when a team is learning a new way of working. Anyone who has to do it all on top of their job goes back to the old way as soon as things get busy.

7. Measure how often it is used. The number of licences tells you how much you pay. It tells you nothing about what happens. Check after two weeks, after a month and after three months how many people really use the application, for what part of the work, and where they work around it. If use drops, go and ask why. The answer tells you what to adjust.

Two people at a table by the window, one listening carefully. Asking why someone does not use a tool tells you what to adjust.

How this can play out, with an invented example. Suppose a wholesaler with 120 people wants to use AI to enter incoming orders from emails and PDFs into the order system. Management tells the inside sales team in advance what the goal is, and that nobody loses their job. Two people from inside sales help choose which orders to start with, because they know which customers send unusual formats. The system prepares the order and an employee approves it. After a month it turns out that one colleague does not use the system for a large customer. When asked, it emerges that those orders keep producing errors. That becomes the next thing they fix. This example is hypothetical. It shows how the steps fit together.

Bring in the works council early

Under article 2 of the Dutch Works Councils Act (Wet op de ondernemingsraden, WOR), an employer running a company in which as a rule at least 50 people work is required to set up a works council. For most companies in our audience, the works council (OR) is therefore a party to the AI conversation. It can be far more than a formality.

Two articles are worth knowing.

Right of consent, article 27. The employer needs the works council's consent for a decision to adopt, amend or withdraw, among other things, rules on the processing and protection of employees' personal data (paragraph 1, point k), and rules on facilities aimed at or suitable for observing or monitoring the attendance, behaviour or performance of employees (paragraph 1, point l). Note the word "suitable". It also covers systems that can do this, even if that is not why you introduce them. An AI application that tracks who processes what and how fast may fall under it. If you take such a decision without consent, the works council can, under paragraph 5, invoke its nullity in writing within one month. Consent is not required insofar as the matter is already substantively regulated in the collective labour agreement (paragraph 3).

Right of advice, article 25. The works council must be given the opportunity to advise on a proposed decision to introduce or change an important technological facility (paragraph 1, point k). Whether an AI application is important enough depends on what it does and how many people it affects.

Whether your application falls under one of these articles depends on what the system does and which data it uses. This is not legal advice. If in doubt, check with an employment lawyer, and read the legal text yourself through the sources below.

Apart from the law, there is a practical reason to involve the works council early. The council hears concerns from colleagues that never reach you. A works council that helps decide from the start what the system may and may not do can explain those agreements to colleagues, and is more likely to be believed than management. A works council that is handed a finished decision has little choice but to be critical. So come with the question, before the answer is fixed.

What to do this month

Three things, which together take less than a day.

Talk separately with three people who do the work you are considering AI for. Ask which part of their work they would most like to be rid of, and what their biggest concern is if AI is added. Write the answers down.

Put on half a page what you want to achieve with AI and what you will not use it for. Share it with the team, and with the works council if you have one.

Look at a tool you already have, AI or not. How many people really used it last month? If the answer is disappointing, you know where to start asking.

Where the Capability Sprint fits

Many of the steps above come together in the way a first application is built. The Capability Sprint is the build phase in which Radical delivers the first working AI application together with your own people. The aim is the handover: your team can then run and adjust it themselves, without the outside party.

Who takes part? According to the description: the people who know the process and will keep working with it afterwards, plus whoever owns the decision. Sending only IT undermines the goal, because the knowledge then lands where the process is not. That matches step 2: the people who do the work help build.

What is delivered: a working application in production, your own people involved in the build rather than watching, and the way of working recorded so that it survives someone leaving. Where needed, a semantic layer is added on top of your existing data, without touching the core systems. The starting point is human on the loop: the systems remain yours, control remains with you, and a person stays able to overrule the system. The test of success comes after delivery. It is the first change your own team makes to the system without calling anyone.

The sprint is the third step. The free AI-Foto shows where you stand, the AI Readiness Scan determines which opportunity delivers the most and is feasible, and the sprint builds the first item on that roadmap. More about the whole approach is on the AI Capability page.

About this page

This piece was written on 5 October 2026 by the Radical AI team. The text of articles 2, 25 and 27 of the Dutch Works Councils Act was checked on wetten.overheid.nl; the English wording above is our own translation. The figures on attitudes towards AI at work come from the European Commission's announcement of the Eurobarometer of February 2025. The reasons companies give for not using AI come from CBS and are provisional figures for 2025. The five concerns, the three forms of resistance and the seven steps are our own structure. The wholesaler example is hypothetical. The description of the Capability Sprint comes from our own pages. No client data was used. This is not legal advice.