Managers are learning AI too: how to introduce it without pretending to be an expert

A leader doesn’t need to know every tool better than their team

A new specialist shows her manager how she used AI to put together a summary in a few minutes. He doesn’t know the tool she used and is tempted to find a mistake straight away. He used to be the leading expert in the department; now people with less time in the job know more about the technology. This need not be a crisis of authority. Management also means creating conditions in which different people’s skills help achieve a good result. The leader is still responsible for priorities, resources, oversight and a fair division of work.

The problem starts when the leader tries to hide what they don’t know while issuing categorical instructions. “From tomorrow, you’re to use AI for all reports” sounds decisive, but it doesn’t explain what may be sent to an external system or who will check the output. A better start would be: “I don’t yet know everything this tool can do. I’d like us to work out together what it’s good for, where it falls short and who signs off the result.”

How to say “I don’t know” and still take responsibility

Research on leader humility describes behaviours such as acknowledging one’s limitations, appreciating other people’s strengths and being willing to learn. That doesn’t mean the leader should be helpless. Employees do, however, need to know what the manager is still responsible for. A good message combines honesty with a clear task: “I can’t yet judge every prompt, but I’m responsible for making sure no unverified information reaches a client.”

In practice, a manager can ask someone familiar with the tool to give a demonstration, and ask a sceptic to prepare a list of situations in which the output might be misleading. Rather than pretending to be an expert, the manager is organising mutual checks. In the meeting, it is worth separating three statements: “this we already know”, “this we will check” and “this method we’re not allowing for now”. After a few weeks, the team will judge the leader less on knowing every feature and more on the quality of the decisions being made.

You may need the sceptic just as much as the enthusiast

An employee says: “I don’t use it because it gets things wrong too often.” The manager may hear this as resistance to change. Yet critics sometimes have valuable observations about data quality, confidentiality or accountability for the result. That doesn’t mean every bit of scepticism is well founded. It is worth asking for a concrete example of a mistake and the conditions under which the tool could be useful. That way, the conversation is about facts rather than judging people.

Be equally careful with the enthusiast. A quickly generated report can look impressive until someone compares it with the source documents. Set up two roles: someone who tests what the tool can do and someone who checks its limitations. Swap them over in subsequent trials. If the sceptic finds a mistake, thank them publicly. That way, reporting a problem doesn’t become a threat to the reputation of the person who raised it.

Set the boundaries for AI use before the first mistake

The team needs answers to practical questions: what data may be entered, which tasks the tool is approved for, what has to be checked independently, who signs off the result and how to report errors. An instruction to “use it responsibly” is not enough. Internal correspondence may need a simpler check than a document for a client or a staffing decision. The scope of checking should depend on the risk.

The EU AI Act provides for measures to support AI literacy that take into account people’s experience and the context of use. This does not mean that every employee has to obtain the same certificate. Nor should every office chatbot be equated with a high-risk system. Managers should know their organisation’s rules and consult the relevant specialists on anything unclear, rather than speaking with certainty about obligations they have not checked.

Start with one task and learn from the result

A first pilot might involve summarising publicly available material or preparing a working draft of a text that contains no sensitive data. Before the trial, agree what is supposed to improve and how you will check it. For example: “For two weeks, we’ll compare the time it takes to produce a first draft and the number of errors found during verification.” If the time saving turns out to be illusory because correcting errors takes longer than doing the work the usual way, that should be recognised as a finding, not as the employee’s failure.

After the trial, hold a short debrief. Ask: what worked, what was surprising, when was the tool not useful and which rule needs to change? Research on psychological safety suggests that being able to voice doubts encourages learning behaviours. That is not a promise that a single meeting will transform the company’s culture. What matters is the manager’s consistent behaviour, especially when someone brings an inconvenient fact to the table.

Learn alongside people instead of demanding instant enthusiasm

Not everyone starts from the same place. One person needs a short example of how to use the tool; another wants to understand the risks and how to roll back a wrong result. Someone else is worried that learning will come at the expense of their normal duties. A well-organised rollout sets aside time for practice and makes clear which tasks can be dropped during this period. There is no point adding a new task to an already full schedule and then criticising people for a lack of enthusiasm.

A manager might say: “At this stage, we’re not judging who is the most cutting-edge. We’re checking whether the tool helps us achieve a defined result without reducing quality.” This approach also encourages people who don’t want to be seen as AI experts to join the conversation. If the trial succeeds, write down the rule for using the tool. If it doesn’t, write down the conditions that would have to change. Either way, the team learns something more important than how to use a single app: how to test new ways of working sensibly.

A manager’s authority does not depend on pretending to know everything. It depends on a clear goal, honestly acknowledging limitations, safety rules and decisions based on the actual results of trials. It is better to run one rigorous experiment and discuss the mistakes than to impose the technology on everyone under the banner of modernity.

How Empatyzer helps a leader guide the team through learning AI

Imagine a manager whose team knows more about AI than he does. Before a meeting, he can ask Em how to admit his limited knowledge without weakening his accountability for decisions. Em will help him prepare an opening, questions for the enthusiasts and the sceptics, and a response for when a demonstration ends in an error. A conversation about his own working style lets him look at whether a need for control or concern for his authority is getting in the way of drawing on other people’s skills. In the team context, he can also take into account differences in attitudes to experimentation, planning and risk, so as to propose a sensible division of roles during the pilot. Em does not check the technical accuracy of the models on the team’s behalf; instead, Em helps organise communication and collaboration around their use. A manager prepared in this way can say “I don’t know”, point out what will be checked and treat critical comments as a contribution to quality. Micro-lessons help reinforce this way of leading conversations even after the first test is over.