In early July, Lünendonk & Hossenfelder in Frankfurt announced its latest industry study on auditing, to be released in August. The study lists revenue rankings and cites a figure that stands out: 57 percent of the firms surveyed expect a sharp decline in the number of new entrants to the profession by 2030. The F.A.Z. report on this topic includes a sentence that’s easy to overlook. Heike Wieland-Blöse, spokesperson for the Grant Thornton Executive Board and honorary professor in Düsseldorf, says: “Knowledge has become modular.” For employers, what will matter in the future is whether applicants have learned how to think. They can acquire specific methods later on thanks to digital tools.
As a result, the certificate is losing its significance as a selection criterion—and from a source one would not have expected. No educational theory, no debate on competencies, no theoretical framework. A certified public accountant simply states what her industry is looking for.
The question remains: what does this statement require? A transcript could be verified, but it was of little use: it measured perseverance within an institution. Whether someone could get to the bottom of a problem was never reflected in it. “Having learned to think” is a better criterion, but it cannot be verified. If the verifiable hurdle is removed without anything else verifiable taking its place, then demeanor, fluency, connections, and background become the deciding factors—in other words, the old markers of the elite, not the promises of meritocracy.
Of all times, this question is coming up again—especially now that it’s truly crucial for society to place the best people in the right positions. But who are the best people for each position? At a time when every language model can instantly generate texts on command that appear to be the result of thought—coherent, plausible, and carefully worded in all the right places—anyone who wants to recognize thought by its results faces a problem, because those results are now readily available at will.
The usual response to this is that we simply need to use AI critically. This falls short because it considers only the supply side—that is, the AI output. What’s more interesting is why we accept this supply so readily. There is an answer to that. It’s seventy-five years old—and uncomfortable.
The location disappears as the criterion increases
First, let’s look at the issue of simultaneity discussed in the same study.
Those who started out in auditing and consulting learned the business through routine work: researching data, creating spreadsheets, and obtaining confirmations from creditors and debtors. It was tedious work that helped you get to know the field, because you had to process a thousand individual cases before you were allowed to assess your first one on your own. It is precisely this work that machines now handle. But how do you take the next step, since the skill is still required, even as the place where you used to acquire it disappears?
How does this affect people and the way our society is organized, as we used to know it?
The Kindest Authority That Ever Existed
Generative AI can be interpreted as an authoritarian model that appropriates public knowledge and repackages it for profit. Obediently and compliantly, it takes over people’s thinking and thereby colonizes them. But: A language model doesn’t command anything. It merely delivers something that looks finished. Therein lies the effect—one to which one can surrender—or not.
The authority lies in the form: complete sentences, a clear structure, and well-considered qualifications in the very places where an expert would place them. A claim to power would be easier to spot. A text that looks like this has already shifted the burden of proof before you’ve even read it. It sounds so reasonable. And disagreeing takes effort. Agreeing, on the other hand, costs nothing. Once you’ve adopted it, you can pass it on as your own in seconds. How awesome is that?
It’s not mandatory. It’s a perk, and you accept perks—whether in the form of Payback points or a finished text. If you want to—or have to. For example, to get a certificate that you might need to show in the future.
Very little has been learned from this so far. To date, this has been rigorously examined in only two narrow contexts. A randomized trial involving developers tasked with learning a new programming library found that AI support led to poorer conceptual understanding, poorer code reading, and poorer debugging, with no significant gain in productivity on average. An EEG study at MIT found weaker neural connectivity and poorer recall of one’s own arguments during essay writing; however, it involved only eighteen participants per group in the crucial session and is described by the authors themselves as preliminary. Neither study provides evidence for knowledge work in the narrower sense—that is, for analysis and contextualization within a specialized field. Nevertheless, both point in the same direction: those who think for themselves first and then reach for the tool fare better. That is the extent of the current evidence, and it is sufficient to support the suspicion.
Before you get too excited: It was almost exactly the same back in the days of cramming for exams. In other words, the classic “knowledge-transfer” charade.
The Demand Side
Let’s take this a step further and question the conventional narrative. It describes what technology does to us—the easy part, because it doesn’t put anyone on the spot.
The other half was formulated in 1950 by a research group led by Adorno. It investigated why people seek domination. The authoritarian character, as described in the studies, demands clear hierarchies and relief from powerlessness. It rejects introspection and ambiguity, adheres to black-and-white patterns, and submits emotionally to the authorities of its own group. The offer of submission is not merely endured—it is actively sought.
That’s exactly what the AI’s user interface does. The simple answer to a complex question makes things easier, and that’s its whole advantage. It’s enough. In a world that’s become overwhelming enough as it is, having something that always responds immediately and never leaves you with doubts is a very good deal.
That is my interpretation. Research on authoritarianism has never examined linguistic models; it is not familiar with them.
And that’s older than any language model
It would be a mistake to treat this as a characteristic of AI.
Every industry that wants to sell on a large scale targets the same structure. For forty years, the fossil fuel industry has provided the simple answer to the complex energy question. The automotive industry sells freedom—and means a product. The chemical industry sells safety to the agricultural sector, the healthcare industry, and for all of our well-being. And the education industry? It sells a catalog of competencies, because a catalog is easier to buy than an open-ended question. Mainstream thinking prevails everywhere because it connects to a need that was already there. This has little to do with whether it’s actually correct.
In the transition from an industrial society to an information society, artificial intelligence draws on and condenses the available knowledge. The condensation of time and space is how growth arises in this system. You don’t have to like it, but it is the logic within which we operate, and this very condensation is also the potential from which something useful and new can emerge.
This leads to an uncomfortable symmetry. Those who have accepted industrial society—along with its hold on the workforce and the associated social safety nets—for decades have little reason to be outraged by the next step. Outrage over AI is often outrage over the fact that this time, it’s their own class that’s being affected.
Suspicion is also directed inward
Steffen Kampeter, CEO of the BDA, made an observation a few days ago in an interview with Table. Media that cannot be ignored, even if one knows who made it. The balance of power within the DGB is shifting. The public sector is growing, the public-sector share of the economy exceeds 50 percent, and this is strengthening the Verdi labor federation at the expense of the traditional industrial unions, which for decades have managed the balance between companies and employees. His comment on this: The public sector is familiar with high energy costs, Asian competitors, and bankruptcies—but only from the media, not from its own experience.
This includes the trade-off. Kampeter is a lobbyist; his conclusions reflect his association’s agenda: lower social security contributions, relax labor laws. The reference to Helmut Kohl—for whom socialism began when the public-sector share of the economy exceeded 50 percent—is a rhetorical device and remains just that. And his grand narrative, “Work holds this society together,” is itself a simplistic answer to a complex question—especially in a decade in which the volume of paid work is shrinking despite all political appeals, and only about 40 percent of the potential labor force is now employed under standard working conditions.
Both are true at the same time. The observation about being out of touch with the market is not invalidated simply because it comes from an employers’ association. Anyone who dismisses it based on the sender’s address is doing exactly what this text is about: they’re categorizing people by camp and skipping the analysis.
And this is where things get uncomfortable for one’s own side. Authoritarianism isn’t a trait exclusive to others. The rejection of ambiguity, the black-and-white mindset, and the emotional submission to the authorities of one’s own group function just as well in enlightened circles—only with the roles reversed. Those who believe they’re on the right side stop questioning. It’s the same mechanism that makes it so easy to agree with the language model.
What this implies, and why it is suspicious
The obvious conclusion is this: become aware of your own authoritarian tendencies, use the machine as a counterpart rather than an oracle, and rely on your own reason. Kant wrote this down in 1784, and it’s still true today.
“Enlightenment is man’s emergence from his self-imposed immaturity. Immaturity is the inability to use one’s own understanding without the guidance of another.” —Immanuel Kant
I’m writing this sentence. Now, here’s the problem with it.
Adorno was suspicious of precisely this movement, for three reasons, all of which are valid. First, the call reduces a structural problem to an individual one: people are told to give up their comforts, while the conditions that produce them remain untouched. Second, those who claim to know what it truly means to be enlightened place themselves at the top of a new hierarchy. Third, as soon as resistance is formulated as advice and sold in courses, it has long since become part of the system.
A blog post that recommends “think for yourself” is itself part of the education industry, offering a simple answer. There’s no arguing that away—only acknowledging it.
What remains is smaller and more inconvenient than an attitude: a process. It manifests itself in three ways in everyday work. In the order of priority—that is, whether one’s own perspective comes before the tool or after it. In the way questions are posed—that is, whether one demands explanations or accepts results. And in the willingness to leave a contradiction unresolved—one that cannot be resolved—rather than smoothing it over to turn it into a recommendation for action.
So, back to the selection criterion. “Whether someone has learned to think” describes an activity, not a level of achievement. It stops as soon as you stop doing it. It cannot be tested. It can be observed—in others, and with considerably more effort, in oneself.
This text also came across to you as well-organized and plausible. That is not an argument for its accuracy.
Sources
- Mark Fehr: “Deloitte and KPMG: How AI Is Displacing New Graduates,” F.A.Z., July 20, 2026 (Lünendonk Industry Study 2026, quotes from Wieland-Blöse, Schenk, Blum, Regierer)
- Steffen Kampeter in an interview with Stefan Braun: “Work Holds This Society Together,” Table.Media Berlin, July 21, 2026
- Adorno, Frenkel-Brunswik, Levinson & Sanford: *The Authoritarian Personality*, 1950 (German edition: *Studien zum autoritären Charakter*, Suhrkamp 1995)
- Horkheimer & Adorno: Dialectic of Enlightenment, 1944/1969
- Kant: Answer to the Question: What Is Enlightenment?, 1784
- Kosmyna et al. (MIT): Your Brain on ChatGPT / Cognitive Debt, 2025
- Shen & Tamkin (Anthropic): Learning Effects in Dialogic vs. Automated AI Use, 2026
