This article was written by Emmalotte Smit and brings together insights from conferences, research and articles on AI and its impact on organizational culture and teamwork. We explicitly see this perspective as a work in progress. New insights may lead us to sharpen, expand or reconsider parts of it. Do you have a response, an addition or a perspective that you think belongs in this article? Email Emmalotte at emmalotte@dir.nl.
AI is obsessed with efficiency. Faster writing, faster analysis, faster searching, faster planning. And that is of course what makes the technology so attractive. Tasks that used to take employees hours can suddenly be done in minutes. Yet there is a question organizations should ask before the efficiency question: what are we actually optimizing for, and what might we lose along the way?
Because once AI becomes part of how people work within your organization, you are not only changing tasks. You are also embedding ideas about what good work is. About what productive behavior looks like, which knowledge matters, how quickly things should move and what role people still have in the process. That is exactly why AI is not only a technological issue, but also a cultural one.
What are we actually optimizing for?
Digital anthropologist Rahaf Harfoush explained at the Nordic Business Forum that many of our current ideas about productivity are not nearly as neutral as we think. The idea that productivity has to be visible, that at the end of an hour there should be something tangible you can point to, comes from the industrial era, when labor could be linked relatively easily to output. In a factory, that relationship was clear: more units produced meant more production.
Knowledge organizations work very differently. Someone trying to solve a complex client problem may spend an hour without sending a single email. Someone preparing a strategic decision has to weigh options, recognize patterns, explore scenarios, connect information and sometimes simply sit with a problem for a while. That hour may turn out to contain the most valuable work of the entire week.
That makes the arrival of AI interesting. If we are not careful, we will use a new technology to reinforce an old idea of productivity. AI saves time, but that time is then immediately filled with more meetings, more tasks and more output. We become faster, but not necessarily better at working.
Harfoush warns in particular about the loss of deep thinking: the time people need to let a problem settle, explore different options and connect ideas before they act. That capacity is already under pressure because of the enormous amount of information we process every day. We consume more and more, while taking less time to reflect and make sense of it. Under stress and constant stimulation, our ability to think well declines. More information therefore does not automatically lead to more insight.
The question is therefore not only how much time AI saves, but above all what we do with that time. Space for reflection, recovery and concentrated thinking helps people build expertise and truly understand complex issues. Daydreaming and boredom can also be valuable because they create room for creativity and new insights. If we fill every minute that becomes available, the old norm remains intact: being visibly busy is the same as adding value. Long working days, dozens of emails and back to back meetings may look productive, while leaving too little space for better thinking.
You could even ask whether employees are then mainly performing productivity. Long days, dozens of emails and continuous meetings look productive, while the time needed to build expertise, recognize patterns and truly understand complex issues disappears.
That also leads to a subtler cultural question. What message are you sending employees if every email contains a Copilot button that can rewrite the text? Or if every document can automatically be summarized? That is of course useful, but it also normalizes certain behavior. The implicit message becomes: do not spend too much time on this. Do not read everything. Do not take too long to formulate. Keep moving. The problem is not that someone occasionally uses AI to rewrite an email. The problem arises when all the digital signals in an organization reinforce the same norm: faster is better.
The interesting question then becomes: what behavioral norm is this tool promoting? If an organization says craftsmanship, attention and curiosity are important values, but its systems constantly push employees towards speed and superficial output, everyday behavior will eventually outweigh the stated values. Harfoush suggests making deep thinking explicit in performance management, for example by expecting employees to spend 15 percent of their time on it. That makes visible that thinking carefully is also a form of productivity.
What happens to connection when AI takes over more of the work?
That brings us to a second risk of AI. In our search for efficiency, we may also optimize human connection out of the organization. Many things that look inefficient from a process perspective are socially extremely valuable. Walking over to a colleague with a question. Working through a problem together. Calling someone because you know they have experience with it. Staying behind for five minutes after a meeting. In a tightly designed process, these may look like unnecessary detours or delays, but these are often exactly the moments in which connection develops.
This matters because people are constantly picking up signals about whether they matter. Does someone ask for your opinion? Does a colleague come to you for advice? Does someone notice that you are struggling with something? Is your expertise recognized? These may seem like small things, but they are precisely the moments from which people infer whether they belong and whether they are valued.
AI can replace many of those moments. If employees start asking every question to a chatbot instead of a colleague, not only does an inefficient step disappear, but so does a moment of contact. Workday research published in Fast Company found that 43 percent of employees say they are more concerned that AI will reduce interaction with colleagues than that AI will eliminate their jobs completely. At the same time, a significant proportion of employees already report having very few conversations with colleagues that go beyond transactional work tasks.
That is why, when implementing AI, you should not only ask which work might disappear, but also which human interactions disappear with it. In our book Micromomenten, we already wrote about the gradual disappearance of the human link in helpdesks. From an efficiency perspective, helpdesks are an obvious candidate for automation: many repetitive questions, a lot of time and high costs. But it is also exactly the moment when someone is stuck and needs help. And that applies not only to customers, but also to employees. An employee who used to call an IT colleague with a question may now get an answer from AI within three seconds. That is faster, but it also means people need each other less often. And that touches something fundamental: the feeling that you are useful and can mean something to someone else.
That connects to the warning computer scientist Roman Yampolskiy links to the concept of ikigai: if machines can perform more and more activities more efficiently, we have to be careful that people do not also lose the sense that they are needed. That is not an argument against AI, but it is an argument for thinking carefully about what you deliberately want to keep human.
At the same time, AI does not have to weaken connection at all. The same technology can be used to strengthen it. A scheduling assistant can fill your week as efficiently as possible, but it can also protect time for deep thinking, reserve space for recovery and deliberately organize meetings with colleagues you would otherwise rarely or never speak to. AI can connect people from different functions or make knowledge and talent visible that would otherwise remain hidden in silos. The difference is not the technology itself, but what you choose to optimize it for.
Do you let AI strengthen your mission, or determine what becomes important?
That brings us to a third point. In the current AI race, organizations sometimes risk forgetting what technology is actually for. Organizations have spent years developing missions, values and strategies. But as soon as a new AI application appears, the question suddenly becomes: where else can we use this? That is a strange reversal. The mission should guide the technology, not the other way around.
Harfoush warns precisely about this FOMO dynamic. If the time that AI frees up is only used for more busyness, more meetings and more output, without creating additional space to understand customers, understand the market or serve the mission better, then speed has become the goal instead of a means.
This is especially relevant for organizations that value diversity of perspectives. Many organizations have deliberately tried in recent years to bring more different voices into decision making. Homogeneous groups are more likely to confirm one another’s assumptions and overlook alternatives. But generative AI introduces something interesting: one question often produces one confidently worded answer. That can easily create the impression that there is one best or objective outcome.
Your organization may say that different perspectives matter, while everyday working habits gradually shift towards treating one answer as the starting point rather than actively exploring multiple interpretations.
There is a risk in that. AI is based on existing data and existing patterns. In that sense, it is inherently conservative: it predicts what is likely based on what has already happened. That makes it powerful, but also limiting. People can do things that fall outside existing patterns.
This became clear at the Nordic Business Forum in the story of Nando Parrado, who survived the 1972 plane crash in the Andes. He casually remarked that if he had asked an AI system at the time about their chances of survival, it would probably have concluded that they were already as good as dead. Yet they survived, because people are capable of extraordinary things, including things that fall outside what appears statistically likely.
That is exactly what you do not want to lose: human creativity and ingenuity. AI can recognize patterns, process enormous amounts of information and perform routine tasks at great speed. Use that power to give people more time for the things in which they add unique value: making new connections, experimenting, trying something that has not been done before and developing solutions that do not automatically follow from existing data.
The time gained should then be given back in the form of contact with others, reflection and moments in which the brain is not constantly under pressure. That is where new insights and better judgements emerge. The real promise of AI is therefore not that people have to think less, but that they gain more time for the kind of thinking, connecting and decision making that is becoming increasingly important.
What does AI use tell you about psychological safety in your organization?
AI can also reveal something about psychological safety in an organization. Amy Edmondson and Jayshree Seth describe in a recent Harvard Business Review article that AI currently has a particular, but probably temporary, advantage: people still tend to perceive AI as a relatively neutral messenger. As AI becomes more deeply embedded in organizations, they expect that perceived neutrality to decline.
In organizations, this can sometimes be seen when employees use AI to formulate a concern or counterargument that they do not feel comfortable expressing directly. This might happen because someone is junior, has less status or is working in a second language. AI can then help sharpen and neutralize the wording of an argument, lowering the threshold for bringing it forward. Edmondson and Seth compare this with situations in the past where employees independently set up an anonymous survey. It not only gives you insight into what people think, but also into the fact that they apparently did not feel able to say it directly.
AI is therefore not a solution to an unsafe culture, but it can be a signal of one. If employees need technology to say something they do not feel safe enough to raise directly, that tells you something about the culture. Pay attention to how colleagues use AI in discussions and decision making.
At the same time, Edmondson and Seth see an interesting opportunity for leaders to partly compensate for the influence of their own position. They advise leaders, for example, to deliberately ask AI to formulate the strongest counterarguments to their own plans. Hierarchy can make people more reluctant to directly challenge a leader.
We would take that idea one step further. Not only leaders, but almost every employee can consciously use AI to challenge their own thinking. This is particularly important because many chatbots are naturally helpful and tend to follow the direction you have already taken.
So do not only ask for an answer. Ask questions such as: What are my blind spots? What assumptions am I making? What is the strongest argument against my proposal? How would someone from another discipline look at this? What alternative explanations fit the same data? In this way, you use AI not to confirm what you already thought more quickly, but to broaden your perspective. AI then shifts from being a consensus machine to becoming a challenger.
The interesting question is therefore not only what people do with AI, but why they use it. Do they use it to improve their thinking, or to avoid a conversation that should really take place between people? Ultimately, the human conversation remains necessary. AI can make possible objections visible, but only people can tell you what is really happening in a team. The sequence Edmondson and Seth recommend is therefore telling: think with AI, then listen to people, and only then make a decision.
How will young talent still develop?
A final risk lies in the development of expertise. The tasks that are easiest for AI to take over have traditionally often been the training ground for young employees. By doing a large amount of relatively simple work, they learn not only the rules of the profession, but also develop experience, intuition and a sense of exceptions. At the same time, they absorb the implicit culture of their organization: what matters here, when do we slow down and what do we consider good professional judgement?
That matters because a great deal of knowledge inside organizations has never been formally documented. Procedures tell someone what to do, but experienced employees also know what to pay attention to, when something does not feel right and when it is sensible to slow down. Precisely that implicit knowledge is often missing from the systems on which AI relies.
Economist K. Sudhir gives an example from the financial sector. For junior employees at banks, the traditional training ground included spending years reviewing loan applications. In doing so, they learned not only the technical side of the work, but also when a case deserved extra attention despite meeting all the formal criteria. If AI takes over that work, the training ground has to be redesigned as well. Juniors can, for example, review AI decisions, deliberately look for errors and exceptions, and discuss with experienced colleagues why they would slow down or deviate in particular cases. Assessing and correcting AI can then itself become a new place to learn.
This also helps prevent a paradox of AI adoption. In the short term, you may need less junior capacity, but in the long term you risk your own supply of senior expertise drying up. You still need to invest in people who will eventually have built enough professional knowledge, experience and understanding of organizational norms to progress into senior roles and exercise sound judgement there. The question before transferring work to AI is therefore: how do people currently develop the professional judgement we will later expect from our senior employees? If the answer is by doing exactly this work repeatedly, then you first need to design a new learning path before handing that work over to AI completely.
Conclusion
The debate about AI is still often framed as if we have to choose between two camps. For or against. Optimist or pessimist. Human or machine. We place ourselves in the technology optimistic camp, precisely because we want to focus on the opportunities AI creates to genuinely help people.
That is also why AI is ultimately a cultural question. Every application makes certain behaviors easier and others less natural. The challenge is therefore not to hope that your culture survives while you introduce AI, but to use your culture as the starting point for how you use AI. Do not start with what the technology can do. Start with what you want to strengthen as an organization.
