The promise of AI in the workplace usually sounds something like this: efficiency, automation, freedom from drudgery. The reality is messier. Companies are cutting entry-level jobs faster than they can retrain workers, and the social fabric that holds organizations together is quietly unraveling. That’s the picture emerging from new research and real-world experiments with AI tools in workplaces across Europe.
The Hollowing Out
Bernard Marr, who writes about AI for Forbes and advises large organizations, describes AI as a “genie” that can handle repetitive tasks. But he emphasizes that jobs aren’t being replaced wholesale—they’re being hollowed out. AI takes the parts of work that employees don’t enjoy: form-filling, data analysis, answering the same questions repeatedly. The concern is what remains.
“If I cut those roles, I also cut off my future employees and the ability to completely rethink how you work as an organization,” Marr said. “I see this in a few AI-native companies, but most companies don’t get that.”
The data backs up a cautious approach. MIT found that 95% of organizations aren’t getting return on investment from AI. McKinsey put that figure even starker: only 1% of companies are seeing real returns. Yet CEOs continue to buy Copilot licenses and claim they’re “doing AI,” without any meaningful transformation of how work actually gets done.
The Hidden Cost No One’s Measuring
Ella Hafermalz, an associate professor at Freie Universität Amsterdam, has spent two years studying how employees actually use large language models. What she found should concern every manager: workers are adopting AI tools organically, without approval, and managers have no visibility into what’s happening.
Employees use ChatGPT to brainstorm instead of asking colleagues, to search instead of consulting internal experts, to polish text instead of getting feedback from peers. The convenience is real. The cost is hidden.
“They were happy to not have to find out where their employees are or their colleagues,” Hafermalz said. “You know how it is with hybrid working—you’re never quite sure who’s where. So, easy, convenient, ask ChatGPT.”
This matters because organizations exist to share knowledge, learn from one another, and check the quality and accuracy of work. When AI replaces those interactions, so goes the mentorship pipeline. Junior employees no longer need to bother senior colleagues with “silly questions”—but those silly questions are how institutional knowledge transfers. They’re how relationships form. They’re how newcomers become insiders.
The Klarna Case Study
If there’s a single data point that should make executives pause, it’s Klarna. The Swedish fintech company, led by CEO Sebastian Siemiatkowski, replaced roughly 700 customer service jobs with an OpenAI-powered chatbot in 2023. The move was framed as a cost-cutting triumph.
By December 2024, Klarna was recruiting again. The chatbot’s work was of lower quality. More importantly, customers wanted a human in the loop. They wanted empathy. They wanted to speak to someone who understood context and could make judgment calls beyond the algorithm.
This isn’t an outlier. It’s a pattern. Companies are discovering that AI can handle the easy stuff but struggles when nuance matters. The technology has “spirited quirks,” as Hafermalz put it—extra fingers in images, confident-sounding but wrong code, outputs that look polished but fall apart under scrutiny.
The Diamond Problem
There’s a growing concern about what researchers call “diamond-shaped organizations.” Picture a diamond: a few leaders at the top, a wide middle of experienced workers, and almost no entry-level positions at the bottom. AI can do what junior employees used to do, so why hire them?
The problem: that’s exactly how organizations train the next generation. If you eliminate the bottom of the diamond, you eliminate the pipeline. No entry-level roles means no way for newcomers to learn the business, build relationships, or develop institutional knowledge. The middle becomes a permanent class with no fresh blood coming in.
“If you’re really looking to grow a culture and a company that’s going to remain distinctive, then you do need to pay attention to those entry-level jobs,” Hafermalz said. “How are entry-level people going to meaningfully interact with your more senior-level management when no one wants to bother each other?”
Larry BlackRock’s Larry Fink made an unexpected argument recently: there will be enormous demand for electricians, welders, and plumbers. The post-war emphasis on university degrees for everyone was probably overdone. The irony is that AI can’t do those jobs—but it also can’t do the entry-level cognitive work that used to serve as a stepping stone into industries like law, finance, and media.
What Actually Works
Priya Lakahni, CEO of Century Tech—an AI education company—frames it around three pillars: foundational knowledge, applied learning, and learning agility. The first is non-negotiable. You can’t build judgment if you can only Google everything. The second means solving real problems, not just passing exams. The third is the ability to learn how to learn, because the jobs of 2030 probably don’t exist yet.
Marr’s research points to the same conclusion. In his book “Future Skills,” he identifies twenty skills needed for the future. Only three are technically oriented. The rest—empathy, critical thinking, strategic problem-solving, the ability to continuously relearn—are distinctly human.
The companies getting this right aren’t focused on cutting costs. They’re focused on what humans do best: the creative, relational, strategic work that AI augments but can’t replace. They’re creating tracks into organizations that allow young people to contribute meaningfully while learning.
The Perspective
Here’s what the AI industry doesn’t want to admit: the technology is impressive but overhyped. Most companies deploying it aren’t seeing returns. The jobs aren’t disappearing in the way predicted—they’re being hollowed out, with consequences that are hard to measure but easy to see in eroded teams, lost mentorship, and quality trade-offs that only emerge later.
The question isn’t whether AI can do your team’s tasks. It’s whether your organization can survive without the human connections that make those tasks meaningful in the first place.