People don’t resist change. They resist being changed. AI adoption requires more than tools, training and communication. Push to pull.
What this article covers
It is often said that organisations need to pay more attention to the human side of AI transformation. That observation is right, but it is rarely followed by much practical guidance on what leaders should actually do. The result is often a limited mix of communication, training and change support around a predominantly technical programme.
This article aims to help fill that gap. It sets out a more complete human-centred approach to AI adoption, built around the idea that people are more likely to embrace change when they understand why it matters, can make sense of what it means for them, and have meaningful influence over how it affects their work. It covers seven elements:
- Three human needs at the centre of AI adoption: mastery and autonomy, meaning, and belonging.
- Diagnosing where adoption gets stuck: using ADKAR to identify the barrier, while recognising that it does not fully explain its underlying cause.
- Understanding the psychological response: using the Four Rooms of Change to understand how people are experiencing the transition.
- Seven human and organisational lenses: including identity, status and power, management systems and unwritten rules, and leadership dilemmas.
- A five-stage adoption approach: Align, Diagnose, Experience & Reimagine, Prove & Learn, Reinforce & Embed.
- The leadership role and dialogues: four recurring leadership responsibilities and five serious conversations to take place throughout the journey.
- The ultimate test of adoption: moving beyond usage metrics to determine whether people are independently shaping, improving and sustaining the new way of working.
Together, these elements provide practical hand-holding for leaders seeking to move from AI roll-out to sustained adoption, and from push to pull.
The challenge
Organisations are investing heavily in AI. They select platforms, build use cases, establish governance, train employees, appoint champions and experiment with agents. Yet adoption often remains disappointing. The technology may work. People may have completed the training. Management may have communicated the ambition. But after the initial pilot, usage drops, employees revert to familiar ways of working and the expected benefits fail to materialise.
The problem is that AI adoption is often treated primarily as a technology, skills and communication challenge. AI changes work. And changing work affects people’s autonomy, competence, professional identity, status, relationships, power and sense of meaning. That makes AI different from implementing another enterprise system. AI can take over parts of the thinking, creating, analysing, deciding and communicating that previously constituted the work itself. A technically successful implementation can therefore still fail as an adoption programme.
A useful starting point is the observation commonly attributed to Peter Senge:
“People don’t resist change. They resist being changed.”
People voluntarily make substantial changes all the time. Resistance becomes more likely when someone else determines what must change, why it must change and what the person’s future role will be.
That suggests a different ambition for AI adoption:
Do not persuade people to accept an AI-enabled future designed for them. Enable people to become authors of that future themselves.
The goal is to move from push to pull.
Three human needs at the centre of AI adoption
Underneath the methodology are three basic human needs that can either work in favour of AI adoption, or turn against it, borrowing from Daniel Pink's "Drive": mastery and autonomy, meaning, and belonging.
Mastery and autonomy
People generally want to become better at what they do, accomplish things, develop and make progress. They also value autonomy: having a meaningful say in what they do and how they do it. AI can strongly activate these needs.
One of the most powerful moments in adoption occurs when somebody experiences:
I can do something I couldn’t do before.
AI has suddenly increased that person’s capability.
But the reverse can happen just as easily. AI can make someone feel less competent. It can threaten expertise built up over many years. It can reduce discretion over how work is performed. It can alter professional status. And it can raise uncomfortable questions about whether parts of someone’s role will continue to exist.
Consider an experienced software engineer who has spent years mastering the craft and is suddenly told that the future engineer may increasingly supervise code generated by AI. The question “Am I becoming a reviewer instead of a maker?” should not be dismissed as resistance to technology. It is a legitimate question about mastery, autonomy, status, identity and dignity.
Leaders need to address that honestly.
Sometimes the right conversation sounds more like this:
“Yes, this may materially change something you have spent years becoming very good at. We don’t yet know every implication. We do know that your expertise remains important in helping us determine what the new way of working should become.”
Generic reassurance is unlikely to create trust when people can see that their work really is changing.
AI adoption should instead create early experiences of increased capability and agency. Wherever possible, people should experiment with AI themselves and have meaningful influence over how it changes their work. Agency does not mean that strategically necessary change becomes optional. It means people have genuine influence over how the change takes place and what the resulting work becomes.
Meaning
People want their work to matter. They also seek coherence: they want to understand why change is happening, where it is leading and how they fit into the story. Meaning exists at two interconnected levels.
First, the change itself needs meaning. Why are we doing this? What are we trying to accomplish as an organisation? How does AI help us deliver our purpose, vision and strategy better?
Starting an adoption programme with “AI is going to change our industry” or “we need everyone using Copilot” may create urgency, but it gives people little reason to care. A stronger sequence is:
Purpose → Ambition → What needs to change → What AI now makes possible
Second, the resulting work needs meaning. What will I contribute? Which parts of my work remain important? What becomes possible that was not possible before? How does my changing role contribute to something worthwhile?
Without clear answers, people will construct their own explanation. AI may then become associated with cost cutting, headcount reduction, loss of professional identity or technology for technology’s sake.
Leaders, together with the communications team, must therefore own, listen to and continuously develop the narrative throughout the change. Purpose belongs at the beginning of AI adoption and needs to remain visible throughout it, rather than becoming a paragraph in the communication plan.
Belonging
People experience change socially. They observe what colleagues do, compare experiences, seek reassurance and construct meaning together. They identify with teams, professions and other groups. That can work strongly in favour of adoption. Seeing a trusted colleague solve a real problem with AI can be far more persuasive than another management presentation about what AI might eventually do.
But the same social mechanism can work against adoption.
Groups can reinforce scepticism among themselves. Negative images of other groups can emerge:
- “Management just wants to reduce headcount.”
- “The AI enthusiasts don’t understand real engineering.”
- “Those people in IT make everything more difficult.”
- “The business doesn’t understand the risks.”
Once these perceptions become part of group identity, sending more information rarely solves the problem. Leaders must therefore pay attention to emerging group dynamics and negative group imagery. These signals need to be surfaced early and addressed through dialogue, shared experiences and visible examples of groups helping one another succeed.
AI adoption should therefore not be designed primarily as an individual learning journey. It should become a social process of experimentation, learning and contribution.
This also changes the role of Teams, Slack or similar collaboration environments. They should not simply distribute communications. They can become part of the adoption infrastructure: places where people ask questions, share experiments, surface failures, contribute ideas and build on what colleagues have created.
Diagnosing where adoption gets stuck
A useful model for diagnosing individual adoption is ADKAR:
Awareness → Desire → Knowledge → Ability → Reinforcement
Its strength lies in helping identify where adoption is getting stuck. If Awareness or Desire is low, more technical training is unlikely to solve the problem. If Knowledge or Ability is the barrier, more leadership communication will probably not help either. But there is an important limitation.
ADKAR is strong at identifying where adoption is getting stuck, but weaker at explaining why.
Low Desire, for example, may not be an individual motivation problem. It may be a perfectly rational response to threatened autonomy, status or professional identity. It may reflect poor meaning, group dynamics or an organisational system whose incentives and decision rights still reward the old behaviour.
Treating resistance primarily as a problem residing inside the employee therefore risks solving the wrong problem.
We find it useful to distinguish three questions:
- ADKAR: Where is adoption getting stuck?
- Four Rooms: What psychological state are people experiencing?
- Our human and organisational lenses: What is driving that response, and what needs to change?
Understanding the psychological response: The Four Rooms
Claes Janssen’s Four Rooms of Change, a model applied at IKEA for example, offers a useful psychological lens for understanding why the same intervention can produce very different reactions. It distinguishes four broad states.
Contentment
“My current way of working is fine.”
These people may not be resisting anything. They simply see insufficient reason to change. They need purpose, curiosity and relevant possibilities.
Denial or self-censorship
“AI isn’t really relevant to what I do.”
More PowerPoint slides about AI may achieve little. They need experience, evidence and honest dialogue.
Confusion or conflict
“I understand that things are changing, but what does this mean for me?”
This is where questions of competence, status, professional identity and future roles become tangible. People need safety, skills, agency and an opportunity to redesign their work.
Renewal
“I can see a better way of working.”
The challenge changes again. People need freedom, support, peers and the removal of organisational barriers so that ideas can become real.
The Four Rooms should not be treated as another sequential implementation model. Different people and teams can occupy different rooms at the same time.
That is why diagnosis matters.
Seven lenses for understanding the human and organisational response
ADKAR and the Four Rooms help with diagnosis, but we need to go further if we want to understand what is actually driving behaviour.
We therefore use seven lenses:
1. Mastery & autonomy
Am I becoming more capable, and do I retain meaningful influence over my work?
2. Meaning
Do I understand why the change matters, and will my future work still matter?
3. Belonging
What are the groups I identify with doing, believing and saying?
4. Four Rooms
Where am I psychologically in relation to the change?
5. Identity, status & power
What might I gain or lose as work, expertise and influence shift?
6. Management system & unwritten rules
What formal and informal mechanisms actually shape behaviour?
And in particular, What do people strive for?Who can give it to them?On what basis? These three questions expose the unwritten rules of the game. People may formally be encouraged to adopt AI while the informal organisation rewards entirely different behaviour. Status may come from being the person who personally solves the hardest technical problems. Promotion may depend on individual visibility rather than helping others become more capable.
Managers may say experimentation matters while rewarding predictable short-term output. AI adoption then collides with the real rules through which people gain recognition, opportunity, influence, security and status.
7. Leadership dilemmas
Which competing goods need to be reconciled rather than simply choosing one side? AI creates genuine leadership tensions:
Productivity versus judgement
We want AI to increase output, while people need to retain sufficient understanding to exercise judgement.
Automation versus learning
Automating work increases efficiency, but performing that work may also be how people develop expertise.
Standardisation versus experimentation
Reusable agents and workflows matter, but standardising too quickly can suppress discovery.
Human control versus agent autonomy
Greater agent autonomy can increase speed and capability, while accountability for outcomes still ultimately needs to sit somewhere.
These are leadership and work-design questions that need to be reconciled and translated into actions and behaviours.
A five-stage approach to AI adoption
So what does a human-centred approach look like in practice?
We use five stages:
ALIGN → DIAGNOSE → EXPERIENCE & REIMAGINE → PROVE & LEARN → REINFORCE & EMBED
Stage 1: ALIGN
Establish why and how we want to change
Start with leadership, not training. Leadership first needs to answer some serious questions:
- Why does AI matter to our purpose and strategy?
- What could it enable us to accomplish that we cannot accomplish today?
- What might genuinely change about people’s work?
- What do we know, and what do we not yet know?
- What will leadership decide?
- What are we genuinely prepared to let people shape?
This creates an initial narrative connecting:
Purpose → Strategy → AI possibilities → Implications for work
The final question matters. Inviting people to contribute after management has already determined the answer creates the appearance of participation rather than agency.
Stage 2: DIAGNOSE
Understand where people really are
Before prescribing training, communication or coaching, establish a baseline.
A survey can measure areas such as AI use, confidence, perceived relevance, ADKAR readiness and the three human needs. But surveys should be combined with interviews, workshops, focus groups, management observations and continuous signals from Teams or Slack.
The diagnosis should examine all seven lenses. It should also surface the unwritten rules of the game: What do people actually strive for? Who controls access to recognition, career opportunity, desirable work, influence or status? What behaviours are rewarded in practice? Where do these mechanisms support the new way of working, and where do they reinforce the old one?
The objective is not simply to produce another AI-readiness score. It is to understand why different groups are behaving as they are.
Stage 3: EXPERIENCE & REIMAGINE
Make AI real, then redesign the work
People cannot meaningfully decide what they think about AI without experiencing what it can actually do. Generic demonstrations and training only go so far. People should work with their own tasks, problems, processes and customer needs.
- What takes disproportionate time?
- What repeatedly frustrates you?
- Where are you unable to deliver the quality you would like?
- What would you like to be able to do that you currently cannot?
- What could become possible if your available cognitive capacity increased dramatically?
Then take the next step. Do not merely ask:“Where can we use AI?”
Ask:
“How should we redesign our work now that these capabilities exist?”
A useful set of questions is:
- Eliminate: What should we stop doing altogether?
- Automate: What could AI perform largely independently?
- Augment: Where could humans and AI together produce better outcomes?
- Reinvent: What becomes possible that was not previously feasible?
- Human: What should deliberately remain human?
This is also where leadership dilemmas need to be surfaced deliberately.
- What happens to learning if we automate the junior work?
- Where does human judgement remain essential?
- How much standardisation do we need without killing experimentation?
- How autonomous should agents become?
These choices shape the future of work.
Stage 4: PROVE & LEARN
Turn possibilities into evidence
Ideas now need to survive contact with reality. Teams experiment on real work, with real consequences. Some experiments will work. Others will not. Productivity may initially fall. AI will miss context. Work may need to be redone. Existing governance will get in the way.
Do not hide this. Narrate the messy middle. Make problems, causes, fixes and learning visible. Otherwise every failure risks becoming evidence that AI does not work, while polished success stories bear little resemblance to what people actually experience.
Peer learning becomes especially important here. Frontrunners can help colleagues, demonstrate what is possible and spread successful practices. But leadership must also watch for an AI in-group developing whose enthusiasm unintentionally alienates everyone else. The aim is not to create AI heroes. It is to make better ways of working increasingly normal.
Stage 5: REINFORCE & EMBED
Change the system that shapes behaviour. A pilot can survive while it receives executive attention, specialist support and extra resources.
The real test comes afterwards. Successful practices need to become part of the way the organisation operates. That may require changes to:
- roles;
- processes;
- decision rights;
- skills;
- job descriptions;
- performance measures;
- incentives;
- governance;
- technology and data access;
- meetings and reporting routines;
- management behaviour.
Some old practices need to stop. And the unwritten rules of the game need to be revisited. If status, opportunity and recognition still depend on behaviours associated with the old way of working, people will eventually return to them.
The questions therefore become:
- What should people strive for in the future?
- Who should be able to confer recognition, opportunity and influence?
- On what criteria?
- What informal status and power mechanisms must change?
Sustained AI adoption is therefore not only about changing individual behaviour.
It requires changing the management system that continuously shapes that behaviour.
Leadership cannot delegate the difficult conversations
AI adoption requires more from leaders than sponsorship and steering committees. Throughout the journey, leaders carry four recurring responsibilities.
Narrative
What story are people currently telling themselves about this change? Leadership must listen to that story, compare it with people’s lived experience and continuously develop the narrative as the change becomes more concrete.
Group dynamics
Who is moving together? Who is moving apart? What images are different groups developing of one another? Leaders need to build positive coalitions and intervene before negative group imagery becomes entrenched.
Conditions
Do people have the capability, support, honesty and genuine agency required to move forward? Leaders need to create the conditions in which people can learn and contribute while maintaining clear direction and accountability.
System
Which formal mechanisms and unwritten rules still reward the old behaviour? Leaders ultimately need to change the organisation around people, rather than repeatedly asking people to behave differently inside an unchanged system.
Five serious leadership conversations
These responsibilities need to become concrete leadership dialogue. Five conversations matter.
Dialogue 1. Why are we doing this?
Connect purpose, strategy and AI possibilities.
Agree what leadership knows, what remains uncertain, what people can genuinely shape and what initial narrative the organisation will use.
Dialogue 2. What are our people actually telling us?
- Where is adoption getting stuck?
- Which Four Rooms are people occupying?
- What is happening to mastery, autonomy, meaning, belonging, identity, status and power?
- What are the unwritten rules of the game?
- And critically: Does this problem require better communication, or do we actually need to change something?
Look at evidence rather than assumptions.
Dialogue 3. What future of work are we prepared to create?
- What disappears?
- What becomes automated?
- What remains human?
- What happens to professional identity?
- Which leadership dilemmas need to be reconciled?
This is where abstract conversations about AI become concrete organisational choices.
Dialogue 4. What are we learning and what will we back?
- Which experiments produce real evidence?
- What should scale?
- What should stop?
- What obstacles does leadership need to remove?
- What employee-created improvements deserve support?
The narrative now needs to move from possibility to evidence.
Dialogue 5. What must change about the organisation itself?
- What becomes standard?
- What old practices stop?
- What needs to change in roles, decision rights, incentives and management routines?
- Which unwritten rules still reinforce the status quo?
- What should those rules become?
- And does people’s lived experience confirm the story leadership originally told about the change?
The ultimate test of AI adoption
Most organisations will measure AI adoption through indicators such as licences, active users, training completion, use cases, productivity and financial impact. Those measures matter. But they do not tell the whole story.
A more mature test asks whether:
- People understand why the change matters.
- They experience AI as increasing their capability rather than merely threatening their expertise.
- They retain meaningful autonomy over their work.
- Their dignity is preserved when professional identities and roles change.
- They participate in shaping what their future work becomes.
- They learn socially and contribute to the progress of others.
- Leadership reconciles the genuine dilemmas created by AI rather than hiding them.
- The formal management system and unwritten rules support the new behaviour rather than pulling people back towards the old.
- People continue improving the new way of working after the adoption team has left.
Purpose gives people a reason to change. Experience makes the possibilities real.
Participation gives them a voice. Agency allows them to shape their future work.
Experimentation turns ideas into evidence. Leadership changes the system so that the new way of working can endure.
That is the shift from AI roll-out to AI adoption, and ultimately from push to pull.
About the author
Bas Kemme is a fractional Head of Strategy & Transformation and boardroom advisor at IntotheNXT, working with leadership teams on strategy, innovation, culture and AI-enabled change. He is currently developing the thinking in this article into an executive education leadership module titled Leading AI Transformation: Human Behaviour Before Technology, focused on the leadership, behavioural and organisational conditions required for sustained AI adoption.
Related reading: Agentic Transformation: what boards and senior executives need to know now