The question I hear most from executives is about people: “How do I prepare my workforce for AI?”
Behind it are usually deeper worries. Will we have to lay people off? Will our best people leave? Will the culture survive? How do I talk about this without causing panic?
Training and communication matter. In my experience, the conversation also needs to address fear, power and identity. Employees are being asked to try a new way of working while wondering what success will mean for their own place in the organization. A presentation about the technology leaves that question open.
Decide what people will do with the capacity
I have seen AI change the mix of tasks inside a role: less gathering information or producing a first draft, and more checking, interpreting and acting on what is available. The extent varies. A new task mix does not by itself establish a smaller role or a financial saving.
The question leaders need to answer is what people should do with any capacity released. If nobody decides, other activity fills the gap or employees assume the unstated purpose is headcount reduction. Neither response is a good basis for learning a new workflow.
Explain the intended change in the work. A customer-service team might use capacity to spend longer on difficult cases. Analysts might investigate the causes behind a variance rather than assemble the report. Those are choices to make and evaluate, not automatic consequences of buying a tool.
New responsibilities also need recognition: maintaining examples, evaluating output, designing handoffs and investigating mistakes. Some are extensions of work people already do. They need time and clear ownership, rather than being treated as an invisible contribution on top of the old workload.
Help people judge the output
Basic AI literacy should be tied to the role. An employee needs to understand what the tool is allowed to do, what information they may share and how to check a consequential result. The relevant skills differ between an accountant, a marketer and a support specialist.
Judgment, communication and relationships remain responsibilities people carry even when AI assists the work. That is a practical allocation of accountability, not a prediction that a technology will never perform a particular task.
Teach people to recognize missing evidence and to ask for help. A fluent answer can be incomplete. Someone who stops a bad result should be able to explain why and have their concern acted on. If the process rewards acceptance and treats escalation as failure, a training course about critical thinking will have little effect.
Training works better when employees can practice on appropriate examples from their actual work and see how the results will be reviewed. A generic demonstration rarely answers the awkward cases that determine whether they will use the tool again.
Why middle managers resist
Middle managers can resist AI for rational reasons. Much of their value may come from brokering information: translating strategy into tasks, aggregating what their teams do and reporting it upwards, and knowing what is happening across their area. Direct access to information changes that position.
If you do not understand this, the resistance will confuse you. If you address it directly, the same managers can help make the change workable.
Engage them early. Discuss how their contribution changes around coaching, developing people, making judgment calls, handling exceptions and solving new problems. Give them time with the tools before asking them to supervise anyone else's use. Make successful adoption a recognized management contribution.
Be honest about the change. A manager who is expected to remove work from a team also needs to understand what the team is being asked to do next. Reassuring them that nothing changes while redesigning their responsibilities will damage credibility.
Managers need practical support too: how to supervise shared workflows, coach someone through a mistake, judge when the tool is failing and address the emotional side of change. The formal project plan often overlooks that work.
Fear is often rational
Most change advice treats fear of AI as a problem to be solved with communication and training. That is too simple, because fear can be a rational response to real uncertainty. Some people may be worse off after adoption. Pretending otherwise destroys trust. Dismissing concerns as irrational adds condescension to uncertainty.
Create conditions in which people can try something and report what happened. Separate early learning from performance ratings where possible. Share experiments that did not work and explain what changed as a result. A promise that every experiment must succeed makes the feedback less reliable.
Make concrete commitments you can keep. A blanket no-layoff promise may be impossible. A narrower commitment—protected learning time, a retraining route, redeployment consideration or a defined consultation process—can be more useful if it is real. State who can answer questions and how the commitment will be applied.
If people are worried about their jobs, address that directly. Town halls are useful for announcements, but people also need smaller settings where they can ask questions without performing for an audience. Bring sceptics into the process; their concerns may reveal an operational problem that enthusiasm has hidden.
What I learned from a volunteer group
At a financial services company in Zurich, the original ambition was to introduce AI widely and quickly. We began instead with volunteers across several teams who wanted to try the tools in their daily work. The early brief was to experiment and tell us what happened.
The customer-service team found practical uses and developed prompts together. The legal team stopped after the tool made confident-sounding mistakes about Swiss contract law. We took it out of that work rather than ask them to trust it. That is a recollection of the team's experience, not a general conclusion about legal AI.
The finance team split. Junior analysts were enthusiastic and senior analysts felt threatened. The team lead described the intended change as using the tool for the repetitive work so people could spend time on the work requiring their experience. That reframing mattered more than another demonstration.
Use spread through conversations between colleagues. I retain the story because it illustrates different reactions inside the same organization. It does not establish an independently measured productivity return or prove that volunteering is always the right rollout method.
I would also make the learning more explicit now. Without putting early experiments into performance targets, record what people tried, what they checked, which outputs were useful and why they stopped a workflow. Volunteers can give valuable feedback; their experience alone does not represent everyone who will later be expected to use the service.
Move from learning to a supported service
You cannot mandate enthusiasm, but you can create space for useful work. Give volunteers approved access and learning time. Recognize the effort spent helping colleagues. Turn repeated discoveries into maintained guidance rather than leaving them in private chats.
Before expanding, check whether the organization can support the intended workflow. Do people have the skills to review it? Are data access and integration ready? Can managers handle exceptions? Does leadership accept that some uses should stop? Those questions concern the service people will depend on, not a general score for their attitude toward AI.
A fast rollout on paper can be slow in practice if employees work around it. A volunteer phase can also go on too long if it never leads to a decision. Set a point at which the owner will assess the evidence and decide what becomes a supported part of work, what needs another test and what ends.
The leadership responsibility is to make those choices visible. Employees should not have to infer the future of their role from a license invitation.
Listen past the first answer
Before the next initiative, have one honest conversation with the team. Ask, “What are you actually worried about?” Then leave room for the answer. The real concerns often follow the socially acceptable ones.
In my experience, they can be more specific than “I'm afraid of losing my job”: “I'm afraid I won't be good at this new thing.” “I'm afraid the work I've spent twenty years mastering won't matter anymore.” “I'm afraid of looking stupid in front of the people I manage.”
Those fears have different answers. Practice can help with one; a credible account of the changing role may help with another. Some need a difficult conversation about a decision leadership has made. None is answered well by pretending the question was only about training.