Depending on who you ask, either artificial intelligence (AI) is transforming white-collar work and rendering many office jobs obsolete, or we are in a massive speculative bubble that is about to burst. With such conflicting messages coming from AI experts, media and all kinds of commentators on social media, it is becoming increasingly difficult to discern signal from noise and to distinguish where genuine innovation is happening and what is just hype and marketing.
Perhaps one of the most prevalent predictions related to AI has been that coding will be automated during the next 6 to 12 months. In fact, we have been approximately six months away from the end of software development as a career according to leading AI labs since around 2023. Recently these claims have even been extended to much of office work. Knowing how the goalpost keeps moving, it is easy to be skeptical of these claims, although credit must be given to some recent developments in AI. Particularly astonishing has been to see how Claude Code has made it possible for almost anyone to build a functioning small app or website, but even such feats are still far from making software developers obsolete.
As someone who has been conducting research and projects with people and companies from business, medical and computer science backgrounds, I now see a massive divide between what the business side believes AI can do, and what the engineers and doctors see AI can do. Drawing from these experiences, this essay provides a critical commentary on the state of AI as of spring 2026.
Not all industries will change rapidly
One of the premises that I want to challenge is that AI will disrupt every industry. We can now see that in many areas AI may speed up work, whether it is used for building prototype apps or quickly filling mostly accurate numbers and references to a consultant’s slide deck[1]. However, for many industries “mostly accurate” is not enough. Highly regulated environments, such as medical, pharmaceutical or defence will (or should) not be early adopters of AI, as long as hallucinations are as big of an issue as they currently are, and when models fail as much as they do when applied outside of controlled lab environments.
Healthcare is a classic example of a domain, where there has been much hype around the idea of AI replacing doctors in certain areas. But successful clinical adoptions are extremely rare, even with older and more robust AI methods that do not rely on generative AI. For the last two years, I have been studying how the introduction of a prototype AI tool affects the training of radiologists together with computer scientists and medical doctors from the University of Copenhagen. Despite nearly a decade of claims of radiologists being automated away, state of the art AI models, such as the one used in this project, are nowhere near the level of even radiology residents[2], who as a sidenote still drive their car to work each day, despite autonomous cars having also been around the corner for years according to many tech evangelists and CEOs.
Not all change is good
As we race to introduce AI into our work, we should also always evaluate whether the change in our workflows and in the quality of our work is positive. There is increasing evidence of us drifting towards accepting subpar or mediocre AI-generated outputs for the sake of speed, cost reduction or simply due to being forced to adopt AI due to hype. This is particularly evident on the internet, which is currently flooded by low quality AI-generated content, from videos to articles to websites riddled with inaccurate content. Recent studies have shown that so called workslop[3] (low quality AI-generated output masquerading as work) is even affecting many office workers, and especially those who diligently fix the outputs of an enthusiast colleague that has adopted AI.
Another challenge introduced by AI adoption is how it changes skill development as well as the focus of how knowledge is transferred between colleagues. In my previously mentioned radiology study, we noticed that the introduction of a diagnostic AI-tool significantly affected how and what radiology residents discuss with their supervisors. After introducing AI, conversations focused more on spotting and correcting mistakes made by the AI model, rather than on discussing (and from that learning) how the actual process of interpreting an x-ray should go, potentially reducing learning and critical thinking.
As a final note, introducing AI to workflows also might shift the role of humans from experts to verifiers. People who were previously actively involved in producing insights and materials, building on years of hands-on experience, now might suddenly find themselves having become passive verifiers, whose job is to spot and correct mistakes made by AI. This type of work is not only more boring but may introduce further errors due to fatigue. More concerningly, new employees who enter a field without previous experience and who no longer get any hands-on learning might face even bigger challenges in becoming experts.
What lies ahead
While these previous examples were critical, the reality is that AI is here to stay in one form or another. When applied properly, AI can bring tremendous value whether it is used in image recognition or text and code generation. It should be kept in mind though that we are in the early stages of AI as a commodity, and there are valid concerns of enshittification[4] eventually happening to the tools that we are becoming increasingly dependent on. While I would not advise anyone to ignore AI, simultaneously I would not suggest anyone to build their skills or company on top of any single AI service, given how hard it is to predict which AI company will be the market leader, and which ones will even exist 12 months from now.
Sippo Rossi is an assistant professor in business analytics within the subject of information systems science at Hanken School of Economics.
[1] For more information, see Fortune’s article Deloitte was caught using AI in $290,000 report to help the Australian government after a researcher flagged hallucinations
[2] A medical doctor that is still training to become a radiologist
[3] For more information, see Harvard Business Review’s article AI-Generated “Workslop” Is Destroying Productivity
[4] The slow decline in the quality of a service as it becomes focused only on maximising revenue
