🔗 The Decoupling
Six Theses on the Future of Work, Growth and Society in the Age of Artificial Intelligence
When a corporation that ranks among the most profitable industrial enterprises in post-war history cuts 50,000 jobs within 18 months while simultaneously announcing a savings program of 60 billion euros, the temptation is to read it as a crisis. As a management failure perhaps, a missed electric transition, Chinese competition taken seriously too late. All of that is true. And yet it falls short.
VW is not a special case. VW is the first large, visible example of a development that is simultaneously beginning in almost every sector. I call it the decoupling, because no other word more precisely describes what is happening: economic growth and employment, which were structurally linked in the post-war model, are beginning to separate from each other. What this means, what consequences it has, and what must now be decided politically and entrepreneurially, I want to develop in six theses.
Thesis 1: The Decoupling Is Not a Cycle. It Is a Structural Break.
The US economy grew by 4.3 percent in Q3 2025. Simultaneously, the unemployment rate rose to 4.6 percent, the highest level in four years. In classical macroeconomics: growth creates employment. This relationship is currently being structurally interrupted, not cyclically.
The reason is technological. AI reduces the marginal costs of cognitive work toward zero. What previously required a team of specialists, a model now handles in seconds, scalable, without labor costs. Capital is substituting labor not only in manufacturing, which has been happening since the 19th century, but in knowledge work, which was previously considered structurally resistant to automation. Accountants, clerks, analysts, lawyers, copywriters: the same logic everywhere. The industries differ. The mechanism is identical.
McKinsey and the Stifterverband estimate that by 2030, around 30 percent of working hours in Germany can be automated, which would mean up to three million career changes. The WEF puts globally displaced jobs by 2030 at 92 million. Whether and to what extent new activities will emerge to close this gap is, frankly, unknown. Model calculations that project new professional fields and promise a positive difference systematically underestimate that retraining speed, geographic distribution, and qualification requirements in the lived labor market rarely work as they do in spreadsheets.
What the current debate still largely ignores: the first wave of decoupling hits white-collar jobs. This is historically unusual, it was almost always the other way around. But it is only the first wave. In parallel, robotics is developing at a speed that has barely registered in public perception. Humanoid robots capable of performing physical work in structured environments are not yet market-ready. When exactly this tips is open. That it tips is not. The decisive mechanism, rarely mentioned in the debate: once a networked robot masters a task, that knowledge is instantly available to every other unit in the same network. What requires years of training for human workers is a software update for a connected robot fleet. This is no longer a linear learning process. This is scaling. Combined with further improved motion control and widespread connectivity, the second wave in logistics, manufacturing, and agriculture could unfold not slower than the first, but faster. The idea of fleeing into trades because there is no automation there is a thinking error that only works on the wrong time horizon.
> The first wave hits white-collar jobs. The second, with robotics, comes for blue-collar. And it could unfold faster than the first.
There is a particular irony in the current situation. While companies and politics intensely debate part-time work, work-life balance, and workplace equality, the majority of decision-makers and employees have not yet asked the real question: What does AI concretely mean for my work, my industry, my profession? This non-engagement is not a neutral position. It is a decision with consequences. Those who ignore the dynamic will be overrun by it, not because they are worse than others, but because skill advantages in this area build up quickly and erode slowly. Ignorance is punished here, not as a moral judgment, but as an economic consequence.
This also has a structural dimension. Some of the loudest societal debates of recent years risk losing their foundation through decoupling before they are concluded. The question of fair distribution of paid work, who may and should work how many hours, loses urgency when paid work itself shrinks. Debates about advancement through education assume that the professional profiles being trained for will still exist in ten years. The demand for more women in middle management positions meets a corporate reality in which precisely this middle management is being systematically thinned out through AI-supported coordination. This does not mean these questions are irrelevant. It means they are being posed within a framework that is changing faster than the debates themselves.
Thesis 2: AI Has No Fixed State. It Is a Moving Target. And the Speed Is What Is Truly Unsettling.
The starting point is known but rarely taken seriously: November 2022. OpenAI releases ChatGPT based on GPT-3.5. For the first time, a large language model becomes mass-capable, freely accessible, usable without technical expertise. What had been reserved for research labs lands within weeks in a hundred million households. That was the first rupture.
What has happened since can hardly be described in the usual language of technological change. We are accustomed to thinking of change cycles in years. A new software generation needs 18 months of development, two years of market penetration, five years to broad adoption. That is the pace at which people, companies, and institutions plan. It is no longer the pace at which AI develops.
Since early 2025, we are experiencing a second, qualitatively different acceleration. Agentic AI, models that not only respond but independently act, plan, write code, detect and correct errors, has fundamentally changed AI's radius of impact. Agentic systems today enable complex knowledge work in a fraction of the previous time, no longer as support for an expert, but as an independent process that takes over tasks from specification to delivery. What this means for industries based on qualified knowledge work cannot yet be fully estimated. That is precisely the problem.
Because AI is still error-prone today. It hallucinates, produces plausible-sounding errors, needs human oversight. Anyone who therefore concludes that the effects are overestimated is making a methodological error: they are evaluating a moving target with a snapshot. There are real factors that could limit the pace: regulation, computing capacity, energy costs, geopolitical restrictions in chip supply. These factors are not trivial. But they address the tempo, not the direction. The relevant question is not what AI can do today. The relevant question is what AI can do in 18 months, and all available indicators point to: more.
> The relevant question is not what AI can do today. They address the tempo, not the direction. And the direction is clear.
Thesis 3: Value Creation Is Happening. Just Differently.
It would be wrong to claim AI destroys value. It concentrates it. AI-driven companies now account for roughly 44 percent of S&P 500 market capitalization, with price-earnings ratios of 31 versus 19 for the overall index. Productivity gains are real. They just no longer flow as wages but as capital returns. That is the actual shift: not less prosperity overall, but a fundamental change in the question of who receives it.
In Germany, according to Bitkom's October 2025 study, 36 percent of companies already use AI, up from 20 percent a year earlier. According to Simon-Kucher, noticeable productivity and employment effects are only expected once penetration reaches 30 to 50 percent. Most companies are approaching this threshold now. What previously stayed below the perception threshold will arrive on the labor market from 2026.
Important here: VW is today the most visible example because industrial layoffs generate headlines. What happens in insurance companies, banks, legal departments, marketing agencies, and newsrooms is structurally identical, just without press conferences. Positions are not being cut. They are simply no longer filled. Hiring freezes generate no reports. The decoupling unfolds quietly while public debate still waits for the big announcements.
For Germany as an export nation with an industrial backbone, this means: competition increasingly takes place at a level where AI adopters build cost advantages that can no longer be caught up without comparable technology deployment. European companies that transform too slowly risk structural competitive disadvantages. European societies that transform too quickly risk social instability. Between these poles, politics must navigate.
Thesis 4: The Legitimacy Problem of Liberal Democracies Arises Not from Poverty but from Status Loss.
I consider the political dimension at least as significant as the economic one. Democracies are stable when their populations experience the system as legitimate, as one that at least tends to consider their interests. This legitimacy does not hang on constitutional texts. It hangs on the perceived fairness of economic conditions.
Wilkinson and Pickett as well as Case and Deaton have demonstrated in extensive empirical work: relative deprivation, the feeling of falling behind relative to one's own reference frame, is a stronger predictor of political radicalization and health deterioration than absolute poverty. Brexit, Trump, the dynamics of populist movements across Europe: these are not the revolts of the truly poor. They are the revolts of those who fear or have experienced decline. The decoupling creates precisely this group: people with employment but without prospects for advancement; people in professions that will look different tomorrow than today; people who played by the rules and still lose.
A policy that attempts to solve this through transfer payments without answering the question of status, participation, and self-efficacy will not solve the problem. Income is necessary. It is not sufficient.
> These are not the revolts of the truly poor. They are the revolts of those who fear or have experienced decline.
Thesis 5: The Instruments for an Orderly Transition Exist. What Is Missing Is Political Courage for Implementation.
The Danish flexicurity model shows that willingness to change in a population can be very high when the state is experienced as a reliable partner. The key is not the specific instrument but the principle behind it: transformations are accepted when they feel like an improvement for the majority. Or at least like preservation. As soon as people get the feeling that change personally leaves them behind while others profit, acceptance tips into resistance. This is not an irrational reaction. It is the correct situational assessment of people who feel the distribution question in their own lives.
This has a direct political consequence that I consider decisive: societal restructuring only works if it follows a simple principle. No one must be worse off than today. Ideally, those who today vote right out of frustration over economic stagnation and status loss will be better off, not because they hold right-wing convictions, but because they see no other way to be heard. These people are not democracy's problem. They are its warning signal. Whoever shapes the decoupling so that middle-class living standards are maintained and gradually improved removes populist protest's most important fuel. This is not a social-policy luxury. It is the prerequisite for the transformation to succeed at all.
I want to make a distinction here that is almost never made in public debate: the decoupling itself is not the problem. An economy that does not realize productivity gains because it maintains employment for its own sake wastes resources that are needed elsewhere. The problem is the absence of mechanisms that make the resulting surpluses socially useful. Two instruments are in focus: a Universal Basic Income (UBI) and a value-added levy on machines and AI-powered means of production. Bill Gates called for a robot tax as early as 2017. Vinod Khosla proposed in February 2026 to completely exempt the bottom 125 million US taxpayers from income tax.
The counterarguments are not without substance: a value-added levy would slow investment and harm the business location. It will be one of the central economic policy trade-offs of the coming decade.
Three of these I consider urgent: the funding base of the welfare state must be expanded to capital returns and automated value creation, because a system primarily funded through wage contributions erodes proportionally to the wage labor it replaces. Europe must rebalance regulation and innovation speed, because a continent that becomes a net importer of AI value creation cannot solve the distribution question on its own. And activities beyond gainful employment must be institutionally recognized, not as rhetoric, but as pension rights and social status.
Thesis 6: The Decoupling Also Creates Opportunities That Were Previously Structurally Closed.
A complete picture also demands this: the decoupling does not only produce losers. It also changes possibility spaces.
AI democratizes production capacities. A small company can today use analytical and communication capacities that were previously reserved for large corporations. This fundamentally changes competitive structures where market access previously depended primarily on capital endowment. However, the opposite also applies: the infrastructure on which AI runs belongs to very few very large companies. Who benefits depends strongly on whether access to this infrastructure remains open or becomes further concentrated. Democratization is possible. It is not guaranteed.
The real opportunity lies not in fleeing to supposedly safe professions, but in society's renegotiation of what counts as valuable work and how it is compensated. Care, education, community work: these activities were not underfunded because they are worth little, but because the wage system historically evaluated by scarcity and replaceability, not by societal contribution. That productivity gains are emerging precisely now that could make this structural error correctable is not an irony of history. It is the real opportunity of the decoupling.
New activities are indeed emerging, but none that can be described in old professional profiles. The person in demand today is not the one who operates AI, but the one who steers, questions, and develops it further: Who identifies which processes can be automated next? Who recognizes where AI systems produce errors that no model detects itself? Who combines medical judgment with AI diagnostics so that diseases are detected earlier? These are not niche professions. This is the new core competency.
Human in the Loop is often misunderstood in the AI debate as a brake, as a regulatory instrument for when something goes wrong. This falls short. It describes a fundamental role shift: from execution to directional decision-making, from production to curation, from process worker to process designer. AI without human judgment in the loop optimizes for metrics, not for meaning. The ability to determine meaning remains human. And it becomes more valuable, not less.
> AI without human judgment in the loop optimizes for metrics, not for meaning. The ability to determine meaning remains human, and it becomes more valuable, not less.
VW is the hook because it is the first large, publicly visible example of this dynamic. The actual story is broader and deeper. It is about whether societies accept the decoupling as a force of nature or treat it as a politically shapeable process. I consider the second option the only defensible one. Not out of idealism, but because the first endangers the political stability that is the prerequisite for any economic transformation.
Sources
1. Handelsblatt: VW cuts 50,000 jobs by 2030. March 10, 2026.
2. wirtschaftsticker.com / Industriemagazin: VW savings program 60 billion; 20% cost reduction by 2028. February 2026.
3. Bitkom Research: AI usage in German companies: 36% (2025) vs. 20% (2024). October 2025.
4. Simon-Kucher & Partners: European Growth Study 2026. Noticeable AI effects only from 30 to 50% penetration.
5. McKinsey Global Institute / Stifterverband: 30% of working hours automatable by 2030; up to 3 million career changes in Germany. 2025.
6. WEF: Future of Jobs Report 2025. 92M displaced jobs, 170M new roles by 2030.
7. ad-hoc-news: US GDP +4.3% Q3/2025 with simultaneously rising unemployment to 4.6%. December 2025.
8. Financial market data: AI-driven companies approx. 44% S&P 500 market cap; P/E 31 vs. 19 (overall index). 2025.
9. Wilkinson, R. / Pickett, K.: The Spirit Level. 2009.
10. Case, A. / Deaton, A.: Deaths of Despair and the Future of Capitalism. Princeton UP, 2020.
11. OECD: Flexicurity. Lessons from Denmark. Updated 2024.