# When We All Have to Lead We are used to AI helping us with our work. What happens when our task becomes helping AI succeed — and how does that role shape us? - Forfatter: Mikkel Freltoft Krogsholm - Type: Essay - Udgivet: 2026-11-07 - Opdateret: 2026-11-07 - Sprog: en - Emner: ai, arbejde, ledelse, læring - Kanonisk URL: https://mikkelkrogsholm.dk/en/articles/naar-vi-alle-skal-lede/ --- You ask an AI to help with your work. It finds a fact, writes a draft or suggests a solution. You carry on. That is how we have learned to understand the relationship: I do the work, and AI helps me. In [a manifesto from Proton](https://proton.me/blog/ai-manifesto), the company behind Proton Mail, chief technology officer Bart Butler turns that relationship around. The developer's job is no longer to get AI to help write code. The job is to help AI write better code faster. That thought reaches far beyond programming. As AI becomes able to carry out more kinds of tasks, we can rephrase it: Your job is no longer to get AI to help you do your work. Your job is to help AI do the work faster and better. The sentence changes where we stand. Before, we sat with the task and had a helper beside us. Now the helper needs to be able to move forward, and we have to work out what it is missing. That is what managers do, isn't it? ## A promotion without a title A manager can make a good result happen without having written the report or built the product herself. She has made the task clear, found the right people and made sure they could get on with it. When something gets stuck, she looks into why. Missing information calls for a different response from a bad decision. That way of working may now arrive on the desks of people who have never managed anyone. You do not need a department or a corner office. You can be the only person in your business and still spend your day setting work in motion, following up and helping others move forward. Imagine an employee who prepares a summary of the company's sales every month. Until now, she has gathered the figures, corrected errors and written the explanation. AI has helped with the spreadsheet and some of the wording. Now AI is given the task of preparing the summary. The employee has to make sure it can retrieve the figures, knows how to count returned goods and can show where its conclusions come from. When the result looks odd, she discovers that two departments record a return differently. She could correct this month's report by hand. Or she could describe the difference and save the rule, giving AI something better to work from next month. Her experience becomes part of the way the work gets done. That can be an unfamiliar way to be good at your job. She has done fewer of the report's individual parts, but made more things work. Her professional pride has to find its place in a result she made possible without producing all of it herself. In [*The Shared Workshop*](https://mikkelkrogsholm.dk/da/articles/det-faelles-vaerksted/), I wrote about the importance of the tools we give AI access to. The management task also includes building that kind of workshop: knowledge you can find, rules you can understand and ways to see whether the work succeeded. Proton points out that the company itself has to keep up. If code gets written faster but waits just as long for a decision, the gain is limited. Helping AI with the work can therefore take an employee from a specific task to a question about why the whole workflow looks the way it does. There is an exciting possibility in that. More people may gain the means to bring an idea to life and a chance to influence how work is organised. But it also gives us a different everyday life. And everyday life is something we practise through, too. ## Work also works on us When you write an explanation, you sometimes discover that you have not understood what you wanted to explain. The sentence will not come together because the thought has not come together yet. You have to go back to the material. The text gets better, and so does your understanding. So work leaves behind more than a product. Something also happens to the person doing it. That is why I am curious about what it does to us if more of our days start to look like a manager's. Researchers have actually followed people through that transition. In [two studies of new managers](https://doi.org/10.1037/apl0000808), Wen-Dong Li and his colleagues found small increases in what psychologists call conscientiousness: the tendency to be organised, dutiful and goal-oriented. They compared them with similar employees who did not become managers and found a link to the increased demands of the role. A [larger German study](https://doi.org/10.1111/jopy.12719), which followed people before and after they became managers, found a different pattern. Self-esteem rose after the transition, while conscientiousness, among other things, fell. Some differences were present long before the first management title. There is no single managerial personality, then, that the role moulds us into. But the studies give us good reason to take an interest in the relationship between the task we take on and the person we become. We bring our habits into the role. Over time, the role may also matter for those habits. Leading AI is a different experience from being responsible for people. An employee has ambitions of their own, may be under pressure at home and may have good reasons for disagreeing with a decision. That relationship makes demands on a manager that AI does not make in the same way. So we cannot simply transfer the research and expect the same effect. What we can do is start looking at what the new working day asks of us. The employee with the sales summary has to spot unclear rules, explain her intentions and distinguish between an impressive result and a correct one. She is practising something different from when she gathered all the figures herself. ## You can get the result without gaining the understanding An [experiment published in 2026](https://www.anthropic.com/research/AI-assistance-coding-skills) makes the difference very concrete. Researchers at Anthropic asked 52 software developers to solve tasks using a set of programming tools they had not used before. Some were given access to AI; others had to work without it. Afterwards, the participants were tested on what they had understood. The group with AI scored an average of 50 per cent of the available points. The group without it scored 67 per cent. Nor was the AI group statistically significantly faster at the task. What was interesting also lay in how the participants used the help. Some handed the work over to AI. Others asked for explanations, asked questions about the concepts or examined the code they had been given. In the researchers' subsequent review, the latter did better. The small groups show different ways of working rather than a ready-made recipe for learning with AI. The experiment measured understanding shortly after a new programming task. That distinction is worth holding on to when we talk about becoming managers: a finished report tells us something about the report. It does not, in itself, tell us what the person responsible has learned. We can get AI to provide an explanation and read it with a nod. Or we can stop at what we do not understand, ask for an example and investigate whether the explanation holds up. Both paths may end with the same text on the screen. They may leave us with different levels of understanding. That fits with a [survey of 319 knowledge workers](https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/). Participants described how their critical thinking with AI shifted towards checking information, bringing answers together and managing the task. Greater trust in AI was also associated with less self-reported critical thinking. In other words, management can be demanding work aimed at understanding. But an approval button can also make something look like management even when you are merely letting a result pass. In [*When Code Is No Longer the Language of Humans*](https://mikkelkrogsholm.dk/da/articles/naar-kode-ikke-laengere-er-menneskets-sprog/), I explored how the machine's work can become understandable to us. Now the question comes back to the human: how do we use that access to actually understand more? ## The one who is also being trained Return to the employee with the sales summary. She asks AI to explain why a figure looks wrong. The answer leads her to discover the difference between the departments' rules. She gives a better instruction and sees how it changes the result. Along the way, she has also gained a better grasp of the company herself. She learns through the work she leads. AI can take care of the painstaking gathering of figures while giving her a chance to explore more connections than she previously had time for. Management need not be a step away from learning. It can give us other things to learn from. If that division of work spreads, the ability to set a direction will be something far more people need to develop. Wanting a result is easy. Explaining what would make it good can require you to discover what you actually mean. That discovery can happen in a conversation with AI, just as it can when you write a difficult sentence. You get something back that makes your own lack of clarity visible. Then you have to keep thinking. What matters is whether the conversation makes room for that work or simply hurries towards a delivery. In [*Education Is No Safe Haven*](https://mikkelkrogsholm.dk/da/articles/dannelse-er-ikke-et-helle/), I wrote about why humanity's place cannot rest on always being on top. The manager's role is no guarantee that we will remain the most capable, either. We can set a direction and still need to learn from the one we lead. That makes Proton's reversed division of work even more interesting. We help AI become able to do the work. The way we do it also becomes part of our own training. So when, at the end of a working day, we ask what AI has got done, there is another question worth taking home: what have we ourselves got better at today?