According to ZDNet, MIT published a study last week titled “Project Iceberg” that used a large-scale computer simulation on the Frontier supercomputer to model the US labor force. The research found that current AI systems can already replace 11.7% of the US workforce, which translates to about $1.2 trillion worth of labor in the country’s $9.4 trillion total labor market. The simulation modeled 151 million workers across 3,000 counties and 32,000 individual skills to create a detailed map of automatable tasks. The study warns that current AI adoption strategies focus on just 2.2% of jobs, mostly in tech, creating a fivefold underestimation of exposure. State officials from North Carolina, Utah, and Tennessee, who co-authored the report, are already planning to use its findings to shape policy.
The real iceberg is white-collar work
Here’s the thing that really jumps out: we’ve been looking in the wrong place. The mainstream panic has been all about AI coming for coders and Silicon Valley jobs. But this simulation suggests the much bigger, submerged part of the iceberg is in everyday white-collar roles everywhere. We’re talking about HR, finance, office administration—jobs that blend routine data processing with some human interaction. These aren’t concentrated in coastal hubs; they’re in every single state. So the narrative that AI disruption is a “tech industry” problem is, according to this research, dangerously misleading. It means a financial analyst in Michigan or an admin coordinator in Ohio might be looking at more immediate automation pressure than a software engineer in San Francisco. That flips the whole script on workforce preparation.
Why a one-size-fits-all approach fails
This is where it gets messy for businesses and policymakers. The study’s core argument is that you can’t just copy the AI playbook from Big Tech and apply it to a manufacturing firm in the Rust Belt or an insurance company in the Midwest. The fabric of each local economy is unique. That’s why the researchers built this as a “virtual sandbox”—they want officials to run their own scenarios. You can see this in action already: Tennessee’s AI Advisory Council action plan explicitly says it will use Project Iceberg to understand local impact. For companies, this complexity basically negates a simple, top-down upskilling mandate. What an employee in a logistics firm needs to learn is wildly different from what a worker in a hospital billing department needs. It’s not just about learning ChatGPT prompts; it’s about understanding which specific tasks in a *specific* role are vulnerable.
Transformation vs. replacement and economic ripples
Now, it’s crucial to pair this with other recent findings. Anthropic’s study suggests AI can make workers 80% faster at some tasks, potentially doubling economic growth. Indeed’s research says AI is more likely to transform jobs than erase them. So is the MIT study being alarmist? Not really. It’s providing the granular, task-level map that those broader predictions lack. The takeaway isn’t necessarily that 11.7% of people will be fired tomorrow. It’s that 11.7% of the *tasks* that make up those jobs are automatable *right now* with existing tech. That’s what leads to transformation—and yes, some replacement. The economic ripple effect is what Gartner calls coming “jobs chaos,” as every business leader scrambles to reconfigure roles based on what the machines can suddenly do. And this chaos won’t be evenly distributed. States that think they’re safe because they have a small tech sector are, as the authors put it, “vulnerable when adoption accelerates in white-collar work.”
What comes next
So where does this leave us? The full report and the Project Iceberg website are meant to be tools. The real work is just beginning. For businesses, the imperative is to audit tasks, not just job titles. For workers, it’s about understanding the *specific* data-centric parts of their role that are ripe for automation. And for policymakers? They’ve got to move beyond generic “future of work” panels and get hyper-local. The study’s value is in its frightening specificity—it moves us from philosophical debates about AI to a concrete, county-by-county analysis of exposure. That’s a much harder, but more necessary, conversation to have. Because if the bulk of the disruption is hidden beneath the surface, we’re all going to need a better map.
