The AI Adoption Power Struggle

AI does not only change tasks. It can change who controls information, who is seen as the expert and who gets to decide. When those shifts threaten status or future relevance, resistance is not always a training problem.

By Glen Maguire, Founder, Matrix AI Consulting

Most people know Vilfredo Pareto for the 80/20 rule. Less familiar is another part of his thinking: power does not remain permanently with the same people. As the sources of advantage change, influence shifts with them. Pareto called this the circulation of elites.

AI is creating a much smaller version of that dynamic inside organisations. It is changing who has access to information, who can produce credible first-pass expertise, who coordinates work and, increasingly, who gets to make decisions.

AI adoption can create a power problem. When AI changes who controls information, who is seen as the expert and who gets to decide, it can also change how people perceive their status and future value.

That matters because AI adoption is usually managed as a technology and skills problem. Give people access. Explain the benefits. Train them. Measure adoption.

But what if the person understands the technology perfectly well and still has a rational reason not to accelerate it?

Your managers may understand AI. That does not mean they want it.

A 2026 preregistered experiment with 2,000 managers in the United States and United Kingdom, by Merrit, Dominski and Hoy, provides one of the clearest signals yet. Managers were randomly shown evidence about AI's productivity benefits, its potential to displace labour, or neutral information.

Their staffing intentions also fell, although by less.

This does not prove managers are systematically blocking enterprise AI programmes. The study measured intentions under an experimental framing, not realised behaviour inside organisations. But it provides causal evidence that, at least in stated intentions, when AI appears to threaten jobs or existing structures, managers can become less willing to support its adoption.

AI changes more than the work

Imagine a manager whose organisational value has historically come from three things. They know more than the people below them. Information flows through them. And important decisions require their involvement.

Now introduce AI. Employees can interrogate company information directly. A junior analyst can produce a credible first-pass analysis in minutes. Meeting summaries, reporting, workflow coordination and follow-through increasingly happen automatically.

The manager has not disappeared. But some of the things that previously made that manager important have become less scarce. That is a very different change impact from learning a new software interface.

The evidence is still young. We should not leap from it to the claim that AI is systematically eliminating middle management. But the mechanism is credible enough that organisations should stop assuming resistance means people "don't get AI".

Sometimes they get it very well.

Awareness is not Desire

This is where change management becomes important. Prosci's ADKAR model separates Awareness from Desire for a reason. Awareness is understanding why the change is necessary. Desire is the personal decision to support and participate in it. More awareness does not automatically create more desire.

That distinction is particularly important with AI. A manager can hold two beliefs at once:

Another presentation about the benefits of AI will not resolve that tension. Neither will another prompting workshop.

This is why effective AI change management has to look beyond tool adoption and ask what the technology changes about roles, workflows, authority and identity. For each significant AI use case, I would ask what work moves, but also:

• Whose status changes?
• Who gains or loses decision authority?
• What expertise becomes less scarce?
• What remains valuable about the human role afterwards?

Those are harder questions. They are also much closer to the real change.

Resistance is not always irrational

There is a trap here. Once leaders recognise that status can drive resistance, it becomes tempting to label anyone slowing an AI initiative as defensive. That would be a mistake.

A manager saying an AI system is not reliable enough for a high-consequence process may be exercising exactly the judgement the organisation needs. Concerns about privacy, accuracy, workload, customer risk and job security can all be legitimate.

Leadership needs to distinguish legitimate challenge from defensive gatekeeping. That requires clarity about decision rights:

• Who can approve a pilot?
• Who decides whether it moves into production?
• Who can stop it?
• Who decides when a workflow or role must change?
• Who accepts responsibility when the evidence is incomplete?

AI is making indecision more expensive

This is the other side of the problem. The answer cannot be endless consultation either. AI capability, employee behaviour and competitive adoption are moving too quickly for organisations to run every decision through months of ambiguous consensus.

The scarce organisational capability is increasingly not awareness of AI. It is having people with enough authority and confidence to make the call when the evidence is incomplete. That means killing weak use cases quickly, backing good ones, setting boundaries, making uncomfortable workforce decisions when necessary, and accepting accountability for the outcome.

Gartner's 2027 CIO research offers a useful indication of how uncommon explicit role redesign still is.

This is survey evidence from an advisory firm rather than causal proof that role redesign produces better AI outcomes. But it reinforces the broader point: organisations are moving faster on AI capability than on the operating model around it.

The lesson is not that every organisation should restructure tomorrow. It is that deploying AI into yesterday's jobs and hoping the organisation adapts around it is unlikely to be enough.

Redesign what makes the role valuable

If AI removes part of the work through which someone historically created value, leaders need to be explicit about what replaces it. Maybe the manager spends less time assembling reports and more time making decisions. Less time coordinating information and more time coaching people. Less time producing first drafts and more time challenging assumptions. Less time supervising routine execution and more time managing exceptions, customers or AI-enabled workflows.

That shift needs to appear in job design, performance expectations, incentives and leadership development, not just in the AI strategy. It is also why executive AI coaching increasingly needs to address leadership, decision-making and organisational redesign rather than simply teaching executives how to use AI tools.

Because the fundamental question is changing. It is no longer only what AI can do.

What becomes more valuable about the person when AI can do more?

Leaders still have to lead

AI adoption creates a difficult balance. Do not ask people to surrender the old basis of their influence without explaining what replaces it. But do not allow uncertainty about status to become an indefinite veto on change either.

When technology is moving this quickly, organisations need both good change management and decisive sponsorship. People need a credible picture of their future role. They also need leaders willing to decide.

AI adoption becomes political when it changes who knows, who decides and who matters. The organisations that treat that as a software rollout will encounter resistance they struggle to explain. Those that redesign the value of the role alongside the workflow have a much better chance of creating something more useful than compliance:

Genuine desire to move with the change.