Two per cent. That is the tiny proportion of companies that have actually cut jobs because an AI solution already in place is taking over the work. By contrast, 60 per cent have already reduced their workforce, simply because they anticipate the future impact of AI. A further 29 per cent have at least put a halt to new recruitment for the same reason. This is revealed by a study by the Harvard Business Review involving more than 1,000 executives worldwide. So, redundancies are hardly being made because of the AI that is actually running within the company. They are being made because of the AI that is expected. It sounds absurd, but it is common practice.
It is therefore hardly surprising that reports are piling up of spectacular U-turns regarding AI transformations that were once announced with great fanfare. GitHub recently provided one reason for this: since June, the Copilot flat-rate subscription has been a thing of the past. Those who work extensively with AI agents now pay for every single token. It is precisely those companies that have plunged deepest into AI automation that are now facing the highest bills. And because they have long since cut back on their workforce, they cannot simply say: ‘never mind then’.
This pattern is evident across all sectors and usually follows the same sequence: first, automate; then, come up against the limits of automation; and finally, bring back precisely the expertise that had been phased out shortly before. For example, in 2024 Klarna replaced parts of its customer service with AI, only to revert, just one year later, to a support model that once again places greater emphasis on human-to-human interaction. Why? Because they had evidently failed to take their customers into account – customers who did not want to rely exclusively on chatbots for advice. In 2026, Ford brought back around 350 experienced engineers after it became clear that automated quality assurance could not, after all, take over the work of the staff who had been made redundant. Two completely different sectors, yet the same lesson can be drawn from their histories: The way back is possible, but it is expensive, slow and not exactly conducive to a company’s reputation. Accordingly, Gartner expects that around half of all companies that have relied on AI for customer service will be rebuilding their workforce by 2027 – albeit on worse terms, with new job titles and without the knowledge that has been lost in the meantime.
What’s more, anyone who cuts back on teams without having operational, tried-and-tested and long-term efficient solutions in place leaves themselves at the mercy of the providers. When prices rise as sharply as they are at present, there are no alternatives. You end up paying whatever is demanded. These are the kinds of dependencies the industry has long been familiar with from proprietary solutions, but now with a completely new intensity and serious financial consequences.
The real problem with the AI economy is therefore not that AI is becoming expensive or reaching its limits. Both are part and parcel of any technological transformation. AI undoubtedly has the potential to improve processes and ease the workload on staff. For this reason, realistic assumptions, robust testing and a plan for dealing with the consequences are urgently needed. Anyone implementing automation should not only calculate cost savings, but also factor in knowledge loss, dependencies and the risk of failure. AI is therefore not a gamble against one’s own workforce, but a tool whose opportunities can be harnessed without ignoring the risks.
This press release is also available at www.pr-com.de/de/avision.
Press contact
Avision GmbH
Christina Karl
Marketing
Bajuwarenring 14
D-82041 Oberhaching
Phone +49-89-623037-967
christina.karl@avision-it.de
PR-COM GmbH
Melissa Gemmrich
Sendlinger-Tor-Platz 6
D-80336 München
Phone +49-89-59997-759
melissa.gemmrich@pr-com.de
