I had a conversation with a manager who showed me his AI-generated summary of the team’s performance. The information was neatly organised, with comparisons and trends suggesting that one employee was falling behind. They were completing fewer tasks, taking longer to respond and showing a downward trajectory on the dashboard. Looking at the report alone, it would have been easy to conclude that there was a performance problem.
As we discussed the findings, I asked what might be behind those numbers. Perhaps the employee had been handling more complicated work or spending time helping colleagues. Conflicting priorities could also have made it harder to finish their own tasks. None of these possibilities should be taken for granted, but they seemed worth exploring before reaching a conclusion. I wanted to know whether he had spoken to the employee about what was happening.
What interested me was how differently a leader could respond to the same information. The summary could help someone prepare for a useful conversation, or it could become the justification for a warning issued before the employee had been heard. Having more information does not necessarily mean we understand a situation better, particularly when we have not considered what the information leaves out.
That conversation captures one of my concerns about AI at work. We are giving leaders increasingly powerful tools without necessarily examining the habits and assumptions they bring to using them. Someone who is already inclined to jump to conclusions may find that an AI-generated report gives those conclusions an appearance of objectivity. A more curious leader might use it to investigate something they would otherwise have missed. The difference matters to the person whose work is being judged.
What happens to the time we save?
There are good reasons to welcome AI into the workplace. Few people would object to spending less time on repetitive administration or searching for information scattered across different systems. The OECD’s surveys of employers and workers, conducted in manufacturing and finance across seven countries, found generally positive perceptions of AI’s effects on performance and working conditions. Those benefits deserve attention alongside the risks.
I have become more cautious about the assumption that saving time will automatically improve the working day. Consider a team that substantially reduces the hours spent preparing weekly reports. Their manager might use some of that capacity to tackle a backlog that has been frustrating everyone for months, or allow people to develop skills they have struggled to find time for. It would be equally possible to raise output targets immediately, without checking whether the team has the capacity to meet them.
In that situation, the reporting task becomes easier, but the overall job may become harder. The OECD’s analysis of AI and job quality identifies concerns about increased work intensity, privacy and, in some cases, reduced human interaction. These findings give leaders a reason to look beyond the hours saved and ask employees how their work has actually changed. A successful implementation should leave room to discuss additional checking, new bottlenecks and expectations that may have grown faster than anyone anticipated.
The temptation to accept a convincing answer
For a busy manager, an AI-generated recommendation can be very appealing. It arrives quickly, reads confidently and appears to bring order to a complicated problem. When several other decisions are waiting, taking the time to question it may feel unnecessary, especially if it confirms what the manager already suspected.
A 2025 study of 319 knowledge workers found that greater confidence in generative AI was associated with less self-reported critical thinking. This was a survey, so it does not establish that AI causes people to lose their thinking skills. It does, however, offer a useful warning about how trust in a tool may affect the effort we put into examining its output.
The implications become particularly important when a recommendation affects someone’s working life. A proposed schedule might meet staffing requirements while overlooking a difficulty an employee has already raised. A performance assessment might count completed tasks without recognising the person who regularly helps colleagues resolve complicated problems. The information may be useful and still be insufficient for the decision in front of us.
I would encourage a leader in that position to consider what they would need to know before explaining the decision to the person affected. If they cannot account for the relevant circumstances, or explain why the recommendation is reasonable, there is more work to do. That might involve checking the underlying information, asking for another perspective or simply having the conversation that the report cannot replace.
Making room for an honest conversation
The same issue arises in communication. Imagine a leader using AI to prepare an announcement about a new way of working. The message is clear and reassuring, but during the meeting an employee asks whether the change will mean fewer jobs or higher targets. Whatever help the leader received with the wording, they now have to respond to a concern that may not have a comfortable answer.
There is nothing wrong with using AI to prepare for that discussion. It could help a leader anticipate questions or notice where an explanation is unclear. The difficulty comes when a polished message creates the impression that concerns have been addressed, even though employees have not had a meaningful opportunity to raise them. Listening involves being prepared to hear something that changes the conversation, including something the leader would rather not discuss.
This becomes especially important when an employee challenges an AI recommendation. A manager may genuinely believe they welcome feedback, yet become defensive when someone questions a system they have supported. Asking the employee to explain what the system has missed is a useful beginning, provided the manager then makes time to investigate. Otherwise, the invitation to speak up is unlikely to mean much the next time a concern arises.
Three things to try this week
These issues can sound like they require a major leadership development programme, but managers can make useful changes in the course of an ordinary week. I would start with decisions and conversations that are already taking place.
Check the context before acting on an assessment. Choose one people-related decision where AI has contributed information or a recommendation. Before proceeding, identify whose perspective is missing and speak to someone who understands the work, ideally the person concerned. The purpose is to establish whether the recommendation holds up once the circumstances are understood.
Ask the team what has become harder. At your next check-in, ask what an AI tool has made easier and where it has introduced extra work. There may be corrections that nobody has accounted for, or new expectations that have quietly become part of the job. Choose an issue you can address and follow up with the team, so the discussion leads to something more useful than a list of frustrations.
Revisit a decision you made with AI. Set aside ten minutes to examine one recommendation you accepted during the week. Consider what you checked, what you took on trust and whether the outcome gave you any reason to reconsider. Over time, this can become a practical way to notice your own habits, particularly where speed or confidence has discouraged a closer look.
What we remain responsible for
I am optimistic about what AI can contribute to working life, but that optimism depends partly on how honestly we examine our own leadership. If a manager tends to rush decisions, avoid difficult conversations or respond to uncertainty by tightening control, a new tool will not automatically change those tendencies. It may give them more ways to act on them, with consequences that reach further than before.
Leadership is not the only factor, of course. Technology design, organisational expectations and safeguards all influence what happens, and individual goodwill cannot compensate for a poorly designed system. Leaders still have an important responsibility within that wider picture: to question what they are being shown, listen to the people affected and remain accountable for the decisions they make.
I come back to that conversation about the performance report because the next step was so ordinary. There was an employee to speak to and a change in their work to understand. AI had helped draw attention to it, but the manager still needed to find out what was going on.
However sophisticated our tools become, we should be careful not to lose the willingness to do that.