Machines at Work now runs weekly in English. Each issue takes one finding from Work After AI, the annual report I write in German, and works it through for an English speaking audience. The full English edition of the report is linked at the end, 109 pages, free.
In June 2026, artificial intelligence was, for the fourth month in a row, the most frequently cited reason for job cuts in the United States. 101,743 positions since the start of the year carried that label. In the same months, the U.S. Census Bureau asked the companies actually using the technology what it had done to their workforce. Around 5 percent reported any effect on headcount at all. Sixteen percent said they had replaced software or equipment. Two percent had fewer employees, and slightly more had hired additional ones.
Both numbers are correct. They describe two different things. One is what companies say when they cut jobs. The other is what happens inside the companies that use the technology. The gap between the two is the subject of the first edition of Work After AI, and it is the reason the title is a claim rather than a question: the great misconception about the AI job killer.
I have held more than 1,000 conversations with owners, boards, and management teams across Germany, Austria, Switzerland, and the rest of Europe. In none of them have I found the job killer. What I have found, everywhere and in the same shape, is a technology that is tested broadly and productive rarely, an entry level that is changing without anyone being able to name the cause, and organizations that do not yet know how to work with a colleague who is not a person. The headline fuses these three things into one story. They have different causes and run on different clocks.
The AI argument serves the balance sheet
The first force is the loudest and it has the least to do with the technology. Since 2023, layoffs at large companies have been announced with an AI justification more and more often, and the pattern is consistent enough to be read. A company that cut 130 positions in July 2026 framed the move as simplifying its structure in the course of its AI build-out. Its revenue had fallen 5.6 percent, its pre-tax profit 3 percent, and the company itself named the Bank of England’s rate cut, not its own technology, as the main cause. The restructuring language, structures simplified, duplications removed, is the classic formula that existed long before generative AI, and it says nowhere which of the eliminated activities a machine actually took over.
The mechanism is not mysterious. A large part of what is presented today as technological modernization is the delayed correction of an overstaffing that arose in the years of cheap money. Venture investors say openly that most large companies are overstaffed by a quarter or more, and one has called AI the silver-bullet excuse. The AI argument is the better argument for a cut that would happen anyway, because a reduction framed as modernization is rewarded by the capital market, while a plain cost cut is not. Personnel costs are the largest freely controllable expense block, so the cut finances the AI investment, and the AI investment supplies the narrative that justifies the cut. That is the whole loop, and the numbers in the Census survey are what it looks like from inside the firms: software gets replaced three times as often as people.
This matters for anyone who owns or prices a company, because the market is currently reading a leaner headcount as evidence of a stronger business, and sometimes it is. Most of the time it is a balance sheet that has been tidied under a technological label, and the two look identical for about four quarters.
The career gap is real, and it is a hiring decision
The second force is the one that is actually measurable, and it is measured at the entry level. Payroll data from the Stanford Digital Economy Lab, in the version from August 2026, shows employment of 22- to 25-year-olds in the most exposed occupations running 19 percent below the level of their peers in less exposed jobs. In the exposed occupations, employment of that age group fell 11 percent since late 2022; in the less exposed ones it rose 10 percent, and the 35- to 49-year-olds in the same exposed occupations also gained 10 percent. The entire gap comes from hiring. Nobody is being laid off in that data. Somebody is not being hired.
Whether the machine, interest rates, or remote work is the cause is genuinely open in the research, and the report says so. What is not open is where the decision sits. The gap is produced in the hiring plan, by companies that treat the junior position as the first line that can be cut when a tool takes over routine work. Two large firms have already gone the other way in 2026. One IT services group took on close to 20,000 graduates last year and plans 25,000 this year, because, in the words of its head of HR, the youngest bring a familiarity with the tools that the experienced levels often lack, and a Big Four firm is replacing its eight-week internship with a paid learning phase of up to a year. The thinned-out junior base and the rising price of experienced staff catch up with everyone else later.
In Europe the drop has so far only been measured in Switzerland, which can mean protection or delay. For the German-speaking region the question turns around entirely. Germany alone will see around 13.3 million people of working age reach retirement by 2040, 30 percent of the workforce. Austria is short roughly 246,000 workers by 2035. A region in that position does not have a job killer problem. It has a problem with companies that stop hiring the people who will be the experienced staff of 2032, and it has that problem while only 0.8 percent of Austrian job postings ask for AI skills at all, the third-lowest share among 22 measured countries.
What the individual gains does not reach the organization
The third force is the consequential one, and it is invisible because it does not compress into a press release. McKinsey’s global survey this year found that 80 percent of respondents report their own productivity has improved with the technology. Thirty-seven percent see any value at all in their company’s operating result, unchanged from the year before. Only around 6 percent have a measurable contribution of at least 5 percent of operating profit, and those 6 percent are a subset of the 37. A year of broader rollout, with 44 percent of firms now scaling across the enterprise, has not narrowed the distance between pilot and production at all.
The small group that does get there differs in one respect from everyone else. Nearly three quarters of them have rebuilt their workflows from the ground up because of the technology, against a quarter of all other respondents. They do not pursue efficiency more often than the rest. The difference is that they additionally target growth and innovation. The technology is the same for everyone. The organization around it is not.
A survey by a software vendor, and one should read it as such, adds the part that boards have not put on their radar. Among decision makers whose organizations are at the earliest stage, using AI as a personal thinking partner, 48 percent admit that their company invests faster than its workforce can learn. Among those at the most advanced stage, where AI runs autonomously inside workflows, the share is 68 percent. The learning gap does not close as an organization matures, it widens. The bottleneck was never the model. It is the speed at which an organization builds the knowledge to use it, and that speed is set by people who were hired, trained, and given permission, not by the next release.
The clock that matters runs in years
Put the three forces on one timeline and the confusion resolves. The AI argument runs in quarters and it is a short, loud episode, peaking now, that will fade when the restructuring cycle has done its work. The career gap runs in years, it began in the spring of 2022 with the rate turn, half a year before ChatGPT, and it is at its deepest today. The long rebuild runs in a decade, and in 2026 its curve is still flat. Carlota Perez described this shape in 2002: every technological revolution has an installation phase driven by finance capital, bubbles included, and a deployment phase in which the gains arrive only once institutions, working practices, and skills have adapted. The job killer belongs to the first half of that story. The questions that decide who wins belong to the second.
That is why the report closes with nine predictions for 2027, each with the measure it will be graded against, and why I will grade them publicly next year, my own and the ones vendors, analysts, and researchers have dated. One of them: the AI argument will not arrive in the German-speaking region in 2027. No company in Germany, Austria, or Switzerland with more than 10,000 employees will justify a cut of more than 1,000 jobs primarily with artificial intelligence, because where a cut has to be negotiated with a works council and priced in a social plan, the technology is a poor excuse. It invites the question which task the machine actually took over.
The decision this leaves in the hands of owners and boards is not whether to buy the technology. Nearly everyone already has. It is whether the organization is being rebuilt around it, starting with the people who are supposed to learn it, at the pace the vendors are already delivering. Those who wait will find in a few years that the rebuild was already under way, in other companies, and they never saw it coming.
The English edition is 109 pages, 135 checked sources, self-financed, bound to no vendor, and free.
Download the report: Work After AI 2026, version 1.1, PDF, 109 pages
The English edition appears in version 1.1, the update the German original received on 18 September, two days after its first publication. It moves the gender gap in generative AI onto Eurostat’s 2025 statistics, which show men in Germany using the tools at work 29 percent more often than women while women aged 16 to 24 are already ahead, and it adds a ninth prediction on that gap. Version 1.2 follows in early November, the closing version 1.3 on 31 December. Every change is logged with its date in the version history, and I will explain it here. A report about a shift that runs in years should not appear once and then stand still for twelve months.
Gerhard Kürner is CEO of 506.ai, the European platform for Service-as-a-Software and agentic engineering. More than 1,000 conversations with boards, owners, and PE funds across DACH and Europe.