In many organizations, the AI conversation begins in roughly the same way: choose the tools, buy the licences, run a few training sessions and encourage people to use them.
A few months later, adoption metrics arrive. How many employees have access? How many use the tools every week? How many hours do they estimate they save? How many presentations, reports or emails do they produce faster?
All of that data is useful. But none of it answers the important question: has the organization changed?
If Maria creates the same presentation in 20 minutes instead of 60, we have a productivity gain. We do not yet have transformation.
If a manager uses AI to summarize a meeting but the meeting still has 12 participants, lasts 90 minutes and produces the same three unclear decisions, the technology has not changed much about the organization. We are simply documenting the same problem more efficiently.
This is one of the important confusions in the current AI conversation: we confuse technology adoption with changing the way work gets done.
People adopt AI faster than organizations change
A McKinsey study published in July 2026 describes three horizons of AI transformation: enablement, automation and reinvention.
In enablement, tools reach employees. In automation, AI enters processes and workflows. Reinvention starts with a different question: how would we design the work if AI had been part of the system from the beginning?
Only 11% of leaders in the research place their organizations in reinvention. Among leaders in reinvention organizations, 48% report enterprise value, compared with 24% in automation and 13% in enablement.
At the enablement horizon, leaders are 5.3 times more likely to report enterprise value when workflows have been redesigned than when they remain unchanged.
The message is fairly simple: AI creates possibilities. The organization decides whether those possibilities become value.
The problem is no longer just 'How do we get people to use AI?'
Training, skills, champions and experimentation still matter. But AI introduces a deeper problem.
If a tool can take over a meaningful part of a role, the question is no longer only how we teach the employee to use it.
We also need to ask: what remains a human responsibility? What can be delegated to AI? What must be checked? Who is accountable when AI is wrong? Which skills become more important? Which activities no longer make sense? Which decisions can move closer to where the information appears?
And perhaps the most uncomfortable question: if we can do the work differently, why keep a process designed for the pre-AI world?
At that point, adoption becomes change management.
Resistance to AI is not necessarily resistance to technology
When people do not adopt a technology quickly enough, the preferred managerial explanation is often simple: resistance to change. It is convenient. It is also sometimes wrong.
An employee can understand the advantages of AI perfectly well and still have rational questions. If I become twice as efficient, do I get more interesting projects or twice as much work? If I automate 30% of what I do, what happens to my role? If AI is wrong, who is accountable?
Those questions are not evidence of an anti-AI mindset. They are questions about trust, professional identity, status, autonomy and security.
McKinsey identifies trust in the organization as critical across all three horizons. Deloitte reports in 2026 that 65% of organizations believe their culture needs to change significantly because of AI, while only 6% of leaders say they are making progress in intentionally designing human-AI interactions.
The technology is moving very quickly. The organization, rather less so.
Training is necessary. But it cannot repair a system that has not changed.
Organizations have a familiar reflex: when a new problem appears, commission a training program.
Sometimes training is exactly the right response. But a work-design problem cannot be solved by training alone.
You can teach a manager to use AI very well. If they still need to approve every step the team takes, the process remains slow. You can teach a team to automate data analysis. If decisions still pass through the same five hierarchical levels, the organization does not automatically become more agile.
If people do not know which data they may use, which risks they may accept, what must be checked and where their autonomy ends, two predictable behaviours appear: some people avoid AI and others use it without saying so.
Both are management problems.
From idea to practice
Want to put the idea into practice?
Discuss an AI adoption intervention for your organizationAI also changes the psychological contract of work
There is a conversation organizations cannot avoid forever: what happens to the time that AI saves?
An employee who saves five hours a week can do more of the same work, take on more complex work, invest in learning, improve customer relationships or experiment.
Or the organization can immediately turn every minute saved into another unit of output. That is tempting. It is also one of the fastest ways to turn a technology associated with autonomy into a new mechanism for work intensification.
Deloitte's UK data already captures the tension: 44% of workers surveyed say they have used AI to automate part of their own job, often without their employer's knowledge.
Sometimes the organization's problem is not that people are failing to adopt AI. It is that they are adopting it before the organization does.
Leaders cannot delegate understanding AI to the IT department
If AI only changes the tools we use, it can be treated as a technology project. If it changes roles, processes, responsibilities, skills, autonomy, power relationships and the way decisions are made, it becomes a leadership problem.
That does not mean every leader needs to become a technical expert. It does mean leaders need enough understanding of the technology to ask better questions about work.
In the automation horizon studied by McKinsey, leaders are 3.9 times more likely to report enterprise value when their leadership teams demonstrate high AI fluency.
Not because the CEO needs to write spectacular prompts. It is simply very difficult to redesign an organization around a technology you understand only through other people's slide decks.
The useful question is not 'Where can we use AI?'
The answer is probably: almost everywhere. That is exactly why it is not the most interesting question.
A more useful question is: 'What would we design differently if we built this process today?'
Perhaps we no longer need the same process. Perhaps a role changes. Perhaps some approvals no longer make sense. Perhaps a decision can move closer to the customer. Perhaps a manager no longer needs to control information, but instead needs to create context and criteria for judgment.
Perhaps human expertise becomes more important precisely because producing an answer becomes extremely cheap. Perhaps people need to learn not only how to get answers from AI, but how to recognize when those answers should not be trusted.
These are no longer questions about software. They are questions about work design.
Adoption is not transformation
We will probably see many impressive AI adoption dashboards: active users, prompts, hours saved, automated processes and lower costs.
Those are useful indicators. But an organization in which everyone uses AI can still operate almost exactly as it did before: the same meetings, approvals, silos, roles and slow decisions. Only the text is produced faster.
Transformation begins when the technology makes us reconsider how work is organized.
That includes processes, skills and roles, but also the things that are harder to put on a dashboard: trust, identity, autonomy, behaviour and the culture in which people are trying to understand what the new technology means for them.
That is why one of the most useful shifts in perspective for leaders is also one of the simplest:
AI adoption is not primarily a technology problem.
It is a change problem.
Sources and context
The data cited comes from research published in 2026 by McKinsey & Company and Deloitte. Percentages and multipliers are reported by the studies' authors for the samples analyzed.
Working with a similar situation?
Start a conversation