When I talk to people about AI, their biggest fear is about jobs. The anxiety is valid and a recent study published by the Massachusetts Institute of Technology (MIT) confirms it, but shifts the perspective on where the next job security risks may emerge.
The paper is called Project Iceberg and was conducted in partnership with Oak Ridge National Laboratory, using one of the world’s most powerful supercomputers. The team simulated interactions between 151 million American workers and over 13,000 AI tools, mapping more than 32,000 distinct skills. Importantly, this is about what the technology can already do today.
The tip of the iceberg
The news constantly talks about IT redundancies, and the research confirms that approximately 2.2% of the American labour market is already exposed in these areas (around $211 billion in wage value). It seems like a lot, but it’s only the tip of the iceberg hence the project’s name.
The real mass is five times larger. Nearly 12% of total wage value ($1.2 trillion) sits in back-office functions: document processing, internal coordination, financial analysis, administration, professional services.
What it measures and what comes next
The MIT researchers emphasise that their index does not predict job losses, does not provide adoption timelines, and does not calculate net effects on employment. The real impact will depend on company strategies, employee adaptation, regulatory decisions, and economic conditions.
What it does offer is a possible projection into the near future (2026–2027) and a map of vulnerabilities, similar to how seismic zones identify exposure without predicting when an earthquake will occur. We know where the risks are, even if we don’t know exactly when they will materialise.
The data suggests that the second major wave of impact will hit back-office functions. This is likely where companies will be tempted to adjust for volume declines in order to maintain growth through cost optimisation.
We expect to see redundancies in these areas starting from 2026 initially in the US, then by 2027-2028 in Romania as well. This depends on several factors, the most important being technological advancement, pressure to implement AI, and the rush to achieve ROI from technology.
Romanian context
For Romania, the context is specific but not fundamentally different. We’re adopting AI slowly, below the European average, but this doesn’t protect us, it just leaves us unprepared. Every company has administrative and reporting functions, precisely where AI’s technical capability already exists.
We’re seeing Romania skip some of the evolutionary stages, with most companies adopting off-the-shelf solutions which means they lose part of the competitive advantage, perhaps unique to this generation.
It’s not about replacing people, at least not in the simplistic sense that usually dominates the conversation. It’s about understanding where ways of working are transforming and how to prepare ourselves to guide this transformation.
The organisations that will have an advantage are those that understand where they’re affected and build the internal capacity to respond strategically.
The employees who will navigate this transformation successfully are those who diversify their skills. IT professionals towards business processes and soft skills; non-technical staff towards technical competencies and adopting AI in their ways of working.
The sooner, the better and ultimately, the line between tech and non-tech (people and companies alike) will become increasingly difficult to draw.