Erosion of Competence Through AI: Why You Don’t Notice What You’re Losing

AI governance and Leadership | article 5

Over the past few decades, we have offloaded a great deal. We’ve handed off mental arithmetic to calculators, spelling to spell-check functions, and our sense of direction to navigation devices. In each instance, it was a gain—things became faster, more convenient, and more reliable. Most of us never even noticed that we had lost something in the process; in any case, no one missed it. Now, we are offloading something else. It is no longer just a calculation or a route, but the very act of grappling with a subject: the analysis, the weighing of options, the formulation of ideas, the process of making an assessment. I consider AI to be one of the most powerful tools ever made available to leaders, and I use it daily. That is precisely why I am raising an uncomfortable question here: what skills am I losing right now, without even realizing it?

What disappears first is not the power of judgment

When people speak of a loss of competence, most think of the obvious: reaching a point where one can no longer do something that one was previously able to do. In reality, however, abilities are interdependent—and the most important of them, one’s own power of judgment, is not the first to disappear.

At the first level lies the craft: analyzing a situation precisely, organizing material, writing in one’s own style, and truly getting to the heart of a subject. This does not yet constitute a judgment; it is a set of tools.

At the second level lies the power of judgment. It does not hover freely above things; rather, it rests upon the first level. Anyone who has never constructed an analysis themselves cannot properly evaluate one created by someone else. They might check it for plausibility—seeing if it sounds coherent or if the figures align—but an analysis can be entirely consistent within itself and yet still be wrong, perhaps because an underlying assumption fails to hold or a crucial variable has been omitted. Neither of these issues will be apparent in the document itself; they can only be discerned by someone who understands the subject matter through their own hands-on work.

At the third level lies that which comes at the highest cost to a company: the ability to differentiate itself. Those who lack their own judgment cannot discern a path of their own; they can only choose between paths suggested to them by others.

The first level supports the second, and the second supports the third. This is why the statement “I’m just having some of my workload taken off my hands” is so difficult to respond to. It is true—yet, at the same time, it can sometimes erode the foundation upon which something entirely different rests.

However, it does not follow that one should delegate nothing to AI. There are tasks that are tedious and require no personal contribution. There are also skills that one can relinquish with a clear conscience—laborious research, for example. The distinction does not lie between delegating and doing things oneself; rather, it lies between delegating consciously and doing so unawares. If you delegate research while fully aware that you will gradually lose that skill, you have made a deliberate choice—a choice that is often the right one. The problem arises when no decision is made at all, and the action becomes nothing more than a reflex.

 

Developing something versus simply rubber-stamping it—there is more to the difference than meets the eye.

I observe this reflex in myself, and I hear it in almost every conversation I have with executives. A report is due, a technical paper, a sensitive email, or a proposal for a decision—the approach is the same: have it drafted, glance over it briefly, and give the green light. The result is ready in minutes, looks polished, and saves half a day’s work.

Formulating is not the tedious part of the work. It is the part in which understanding emerges.

The obvious objection is: “I can just read the finished text, and then I’ll have the knowledge, too.” That’s true—but it’s a different kind of knowledge. When you work through a subject yourself, you make hundreds of small decisions along the way: what belongs together, what should be left out, where a line of reasoning doesn’t quite add up. In the process, you gain not only knowledge about the subject itself, but something even more important: an awareness of your own gaps in understanding. Afterward, you know exactly where you’re on solid ground—and where you aren’t.

Knowledge acquired through reading lacks this texture. It is smooth. You know the answer, but not the uncertainties that lie behind it. And that is precisely what becomes apparent the moment someone asks for further detail—a client, a member of the supervisory board, or a journalist.

With emails, this carries even more weight. The real work has never been about the actual drafting; it has always been about considering how the message will be received—by a disappointed customer, a colleague who feels overlooked, or an employee in a difficult situation. That crucial shift in perspective is lost when the text is already generated for you. This is further compounded by a superficial prompt: if your request is imprecise, it’s a sign that you haven’t thought things through carefully beforehand.

Anyone who merely nods their approval to a draft has taken note of the subject matter—but they have not truly grasped it.

Why no one notices

At this point, the following objection invariably arises: “We said the same thing about pocket calculators—and we didn’t end up any less intelligent.” This objection is valid, but it overlooks two differences. When it comes to mental arithmetic, everyone knows perfectly well that they can no longer do it. And if there’s any doubt, it can be verified in thirty seconds.

Neither applies to the faculty of judgment. In this regard, there is a finding from the field of medicine that I consider the most revealing in this entire debate. Experienced endoscopists—each with more than 2,000 procedures under their belt—worked for several months with the aid of AI. Afterward, their ability to detect abnormalities without such assistance was measured. Their detection rate had fallen from approximately 28 percent to about 22 percent—and this among specialists, not novices.

The real point, however, is not the decline itself. It is that these doctors only became aware of it because someone took measurements. The same applies to pilots, whose manual flying skills are assessed twice a year in a simulator. For an executive, no one takes such measurements. There is no metric for gauging sound judgment, and no simulator for decision-making.

That is why this decay does not make itself known. It waits for the day when it truly matters.

When the draft is already the decision

This becomes most apparent in the context of decision-making. A proposal for a decision has always served two functions: it is meant to inform, while also compelling the person who drafts it to weigh various options, articulate underlying assumptions, and contend with uncertainty. If such a proposal arrives at the table fully formulated and plausibly justified, the second function is lost. In that case, anyone who merely reviews and follows the document is no longer making a decision—they are simply confirming one.

What is missing here can be precisely identified: the history of discarded options. A finished template always presents the rationale for the course of action it recommends. It does not reveal which alternatives were considered and why they were rejected, nor does it show the single underlying assumption upon which the entire structure rests. Anyone who has built a template themselves knows exactly where it is shaky. Someone who has merely read it, however, sees a self-contained argument. Self-contained arguments appear compelling, even when they rest on a weak assumption.

Then there is a second type of gap, which is even harder to detect. An AI can only discard what it has already considered. An option that it never took into account does not even appear in the output as something that was rejected; it is missing without a trace. Only someone who has truly mastered the subject matter can tell when something is missing.

In the past, there was a signal we all relied on: poor-quality work could be identified by its form. It was incomplete, sloppy, or poorly phrased. AI has severed this link between quality and form. Today, an incorrect result can look just as good as a correct one—sometimes even better, precisely because it is not constrained by the need to adhere to limitations. The appearance of diligence is no longer a reliable indicator of quality.

What helps instead are two questions that cannot be answered unless you have thoroughly grasped the text in question: Which option did you discard, and why? And which single assumption would have to be incorrect for the decision to change?

The boundary lies between closeness and distance.

So, where should the line be drawn? A school-based experiment involving around a thousand teenagers provides a surprisingly practical answer. One group worked with an AI tutor that supplied solutions; as long as the tool was available, they performed significantly better on the tasks than the control group. However, in the subsequent exam—conducted without any aids—their performance dropped by about 17 percent. A second group used the same system, but it provided only hints rather than full solutions. This group practiced even more effectively and showed virtually no decline in performance during the exam.

The same material, the same tool—yet the effect is opposite. The difference lies not in the extent of the help, but in its direction.

What matters is not what you give up, but whether it brings you closer to your goal or takes you further away.

Anything that brings you closer is good: conducting research, organizing material, sorting out connections, checking calculations, encountering gaps, or being challenged. Anything that keeps you at a distance is risky: arriving at a pre-packaged conclusion, a ready-made assessment, or a final judgment.

When an entire organization forgets

Up to this point, the focus has been on you. The second part of the problem concerns your company—it unfolds more slowly and is, for that reason, more unpleasant.

Leaders do not need to issue directives for a particular behavior to take hold; it is enough for it to become clear what is being rewarded. In environments where quick, polished results are what count, the organization learns to prioritize exactly that—and no one will ever admit that they did not truly earn their position through their own efforts.

Since no one there measures the capacity for judgment either, this decline remains collectively invisible. A case study from a company that had largely automated its accounting illustrates the mechanism: because errors rarely occurred, employees began to review the results with increasing superficiality—and because they reviewed them less thoroughly, their ability to conduct such reviews continued to diminish. It is a self-reinforcing cycle—one that went unnoticed by both the employees and the managers.

He only becomes visible at two points, and both come late.

The first scenario is the exception. In the field of automation research, it has been known for over forty years that the more reliably a system operates, the less frequently a human needs to intervene—and, consequently, the less prepared that person is when intervention finally becomes necessary. It is precisely at such moments—during a crisis or when a discretionary decision is required—that the capacity for sound judgment is called for.

The second factor is competition. An experiment involving authors showed that AI support made individual texts more creative—but also made all the texts more similar to one another. Applied to businesses, this means that if everyone uses the same systems and no one possesses the foundational expertise to even identify a unique path forward, no competitive advantage can emerge. Instead, you end up with a bland, uniform sameness.

One might counter—and I take this objection seriously—that an organization heavily supported by AI is simply more efficient. Perhaps. But that holds true only as long as nothing goes wrong, and only as long as others aren’t doing the exact same thing. An organization that has lost its capacity for critical cross-checking has neither a fallback option nor anyone to notice that it’s time to have one.

The order is what counts

Three things have proven effective for me. None of them requires giving up AI.

The first point is really just a matter of sequence: form your own assessment first, then consult the AI. If you ask the AI ​​first, you end up with an “anchor” that your own thinking tends to fixate on. If you take a position yourself first, the AI ​​becomes a sparring partner. This leads to what I find to be the most compelling use of all: instead of asking the AI ​​for a solution, you ask it to challenge you. Where is the flaw in my reasoning? Which assumption doesn’t hold up? What arguments are there against my chosen course of action? Used in this way, AI sharpens your judgment rather than replacing it—it’s the difference between a tutor who simply provides answers and one who offers guidance.

The second point is to keep your practical skills sharp through regular practice. In aviation, pilots are required to fly manually on a regular basis—precisely because the autopilot is so reliable. Leaders, however, have neither such mandates nor a flight simulator, and good intentions rarely last. A more realistic approach is to anchor this practice to events that are already on the calendar: a strategy retreat, a decision with significant consequences, or a difficult conversation. In these instances, make a conscious effort to prepare without relying on your usual tools—not as a mere exercise in discipline, but because this is the only scenario in which you would actually notice the difference.

The third point concerns the organization, and this is an area where the wrong approach is often taken. Do not try to determine whether a document or proposal was created using AI. It is impossible to tell, and such an approach fosters a culture of surveillance rather than quality—further driving the use of AI into the shadows. Instead, shift the nature of the conversation around the document. Ask about the options that were discarded and the core assumptions that underpin the work. No one can answer these two questions without having truly grasped the subject matter—regardless of what tools were used to produce the document. This does not penalize the use of AI; rather, it penalizes a lack of deep understanding. And that is precisely what matters.

 

Conclusion: Competence quietly disappears

Competence is not lost with a bang. It slips away quietly—and the insidious thing is that, as it vanishes, it actually feels good. Faster. More assured. Less burdened.

This article is therefore not an argument against AI. I use it daily and wouldn’t want to be without it. Rather, it is a plea to consciously draw a line—a line that no one else will draw for you. Use AI to delve deeper into a subject; do not use it to keep that subject at arm’s length. And make this a deliberate choice, not an automatic reflex.

Ultimately, someone bears responsibility for what the AI ​​has prepared. However, this responsibility is only as meaningful as one’s ability to object when in doubt. Anyone who is no longer able to object can no longer truly fulfill their responsibility; they are merely signing off on things. That is precisely the focus of the next article in this series.

And that brings me to a question that has occupied my own thoughts while working on this text: How would you actually notice if your own judgment were beginning to decline—and who in your organization would tell you? I look forward to our exchange.

What to expect in this series

This series guides you through the key leadership issues surrounding AI in the enterprise—offering practical insights drawn from firsthand experience.

So far, the focus has been on the people in the team. In the next article, I will turn my attention to leaders themselves—and to a danger more serious than being replaced by AI: the gradual erosion of their own judgment. Because anyone who wants to teach others not to trust AI blindly must begin with themselves.

In the previous article, I addressed the question: Implementing AI Means Leading People: Four Fault Lines No Rollout Plan Reveals

You can find all articles in the series and further background information on our topic page. „KI-Governance and Leadership“.

Who this series is relevant for

This series is aimed at managing directors, board members, advisory board members, and executives who do not wish to reduce the approach to AI to a purely IT-related matter. It is particularly relevant for companies where AI is already being used in isolated areas, yet roles, responsibilities, and guardrails remain undefined.
Especially in medium-sized enterprises, this quickly creates a complex mix of opportunities, uncertainty, and pressure to act. That is precisely where leadership is required—not later, but now.

About the autor

Dr. Bernd Kappesser is a partner at empiricus GmbH. With his many years of experience in leadership roles in the IT and technology sectors, he possesses in-depth expertise in transformation, organizational development, and strategic leadership.

He supports executive boards, management boards, advisory boards, and leadership teams in consciously shaping their roles, effectively leading change, and positioning their organizations for the future amid the challenges of digitalization, artificial intelligence, and regulation. His approach combines entrepreneurial practice, strategic thinking, and personal reflection—with a clear focus on sustainable impact.
He guides management teams, executive boards, advisory boards, and leadership teams in consciously shaping their roles, effectively leading change, and positioning their organizations for the future amidst the complex interplay of digitalization, artificial intelligence, and regulation. His approach combines business practice, strategic thinking, and personal reflection—with a clear focus on sustainable impact.

Find out more about our services Executive Advisory und Leadership Development. Get in touch – Contact