The people who hang back longest in AI adoption programmes are rarely the least capable. I have noticed this pattern repeat across teams: AI adoption resistance concentrates in the experienced, the committed, the people who care most about doing good work.
The usual explanation is fear of change, or attachment to the status quo. I think it is simpler and harder than that. What they are protecting is not the tool. It is the basis on which they understand their own contribution.
They have spent years building competence in specific tasks. They learned to do those tasks well. They apply judgment and care to them. Those tasks have become how they understand what they are for. Then AI arrives and handles them fluently, at speed, without apparent effort.
Nobody tells them what remains.
The wrong diagnosis
Most transformation programmes treat resistance as a process problem. The intervention looks familiar: a training plan, tool access, adoption dashboards tracking usage, a change management workstream, a communications campaign about the exciting future of human-AI collaboration.
This addresses the surface behaviour (not using the tools) without touching the cause (the threat to how people understand what they contribute). The dashboard goes green. The behaviour stays cautious.
Joe Hudson, who coaches leadership teams at OpenAI, Anthropic, and Apple, argues that what looks like resistance is often a form of self-protection: a neural response that shuts down experimentation when the risk of failure feels too high. People stop trying the prototype, stop questioning the AI output, stop running the experiment. Not because they cannot. Because something in the environment signals that being wrong here carries a cost they are not willing to pay.
Standard change management does not reach this. It addresses process compliance, not identity threat.
What the experienced person is asking
A 2024 study found professional identity threat to be one of the primary drivers of passive resistance in AI adoption programmes: not sabotage, not overt refusal, but a quiet withholding of effort.
The threat is not about the technology itself. It is about whether there is a credible answer to the question the experienced person is privately asking: if AI can do the thing I built my career doing, what am I for?
The answer is visible in the work that resists automation: judgment applied to novel situations, empathy in interactions where the technically correct answer is the wrong one, care that cannot be procedurally specified. The problem is that these capabilities look invisible precisely because they are exercised on top of the tasks AI has replaced. When the tasks go, people assume the value went with them.
The job of a transformation programme is to make that assumption wrong.
Two parts, not one
What actually works is a combination, and most programmes get only half of it.
The first half is experience: getting people to actually use the tool, practise with it, find out what it does and what it does not do. Not a demo. Real work. The fear of AI often lives in the imagination, and direct practice deflates it in ways that no presentation can. People start to see concretely where the AI is unreliable, where it needs direction, where the call is still theirs.
The second half is harder. It is making explicit what remains uniquely theirs once the repeatable tasks have moved. The judgment about what the AI missed. The empathy for the person whose situation the AI described accurately and understood wrongly. Care for the outcome rather than the output. These are not consolation prizes. They are the things the work was always for, now visible because the scaffolding around them has been cleared.
Neither half works alone. Experience without the reframe leaves people using the tool while privately grieving what they used to do. The reframe without experience stays theoretical: well-intentioned words that do not connect to anything they can actually see in the daily work.
When both land, something shifts. Teams start being comfortable being wrong quickly. They surface difficulties rather than routing round them. They run experiments because they trust the outcome will not define their standing. None of that shows up on an adoption dashboard.
Not just compliance
The demand for human judgment has not diminished as AI output has increased. Someone has to evaluate the outputs, push back on the confident errors, make the calls the AI was not built to make. The supply of people who can do that well is constrained by how many organisations have helped their teams understand that this is now what they are for.
Programmes that fix the process without addressing the identity question produce compliant teams using tools on tasks that do not matter, and avoiding the situations where their judgment is most needed.
The organisations that get this right build something different: people who know what they are for, and who bring that clearly into the work AI cannot do. That is harder to deploy than a licence.
It is also harder to copy.
Leave a Reply