MIT and Oak Ridge National Laboratory just released Project Iceberg, a massive simulation that tracked 151 million US workers across 32,000 skills and 923 occupations to figure out which jobs AI can already automate today.
The findings = AI can technically replace 11.7% of the American workforce right now… affecting $1.2 trillion in wages. That's not a prediction for 2030. That's what's possible with current technology.
Here’s the deal:
- If you only look at where AI is actually being used (mainly computing and tech), it looks like just 2.2% of jobs are affected; about $211 billion in wages.
- But MIT calls this the “Surface Index.”
- Below the surface lurks the real threat: cognitive work in finance, healthcare, administrative roles, and professional services that AI could automate but hasn't yet.
What changed everything: The Model Context Protocol (MCP). I got asked the other day if AI could do my job. My answer was kinda, because it could do all the parts in isolation, but not together.
Think of your job as requiring access to dozens of different tools (or work surfaces): email, databases, spreadsheets, CRMs, project management software, team conversations, meetings, etc. Until recently, AI assistants were stuck outside this ecosystem, unable to actually do anything with your tools.
That changed in late 2024 when Anthropic released the Model Context Protocol. It basically lets any AI model plug into any data source or tool. Before MCP, every AI integration required custom code. Now, AI agents can autonomously access databases, manipulate spreadsheets, query APIs, and coordinate workflows through a standardized connection.
The explosion happened fast. As of March 2025, there are 7,950+ MCP servers available, from both indie developers and major players like Microsoft, OpenAI, and others. This created what MIT calls the “Agentic API Economy”, where AI systems can autonomously perform valuable tasks that previously required human labor. An AI agent can now check your calendar, book a room, send invites, update your project plan, and generate a financial report without human intervention. Put differently: doing your job.
Project Iceberg tracks every one of these MCP servers and maps them against the 32,000+ skills in the American workforce. When a new MCP server launches that automates financial analysis, the Iceberg Index immediately calculates which occupations, industries, and geographic regions just became more vulnerable.
Plot twist: MIT says the biggest impact isn't in Silicon Valley. Rust Belt states like Ohio, Michigan, and Tennessee show modest Surface Index values but massive Iceberg Index numbers. Why? All that cognitive work supporting manufacturing (financial analysis, administrative coordination, professional services) is highly automatable. If you work in these industries / states, you might wanna hide this report from your boss…
Don’t worry though, states are already moving. Tennessee cited the Iceberg Index in its official AI Workforce Action Plan this month. North Carolina's using it to drill down to census-block-level data, identifying which skills in specific zip codes are most at risk. Utah’s already building policy scenarios with the platform.
Here's what makes this different from typical AI doom-and-gloom (a.k.a Anthropic CEO Dario Amodei predictions): Project Iceberg isn't predicting layoffs. It's an early warning system. The interactive tool lets policymakers test different scenarios. Go ahead and try it: you can use it to reallocate workforce dollars, tweak training programs, or adjust tech adoption rates before committing billions to implementation.
So that's what AI can do today. But what about tomorrow?
Ethan Mollick recently shared the most comprehensive expert survey on AI's future, and the results reveal a massive gap between what AI company leaders predict and what 339 top experts actually forecast for the next 15 years.
The Longitudinal Expert AI Panel surveyed superforecasters (people with proven track records of accurate predictions), top computer scientists, economists, and policy experts. These aren't random opinions—participants spent an average of 44 minutes on each survey and submitted over 600,000 words explaining their reasoning.
Here's what the experts predict by 2030:
- Work transformation: 18% of work hours will be AI-assisted, up from 4.1% today—a 4x increase in six years.
- Energy demands: 7% of US electricity will power AI training and deployment (that's 1.5x today's entire data center load).
- Math breakthroughs: 23% of experts think AI will saturate the FrontierMath benchmark, solving problems that typically take PhD students several days.
- Private investment: AI investment will double to $260B annually.
By 2040, things get wilder: experts give a 60% chance AI will substantially help solve a Millennium Prize Problem (some of math's hardest unsolved challenges), and predict 30% of adults will use AI for daily companionship.
Wait—something doesn't add up here.
MIT says 11.7% of jobs are automatable right now. Experts predict 18% of work hours will be AI-assisted by 2030. That's... pretty close, actually. Here's where it gets interesting—the median expert forecasts are way more conservative than what AI company leaders are saying publicly.
Elon Musk claims all jobs will be replaced by 2030. Dario Amodei predicts 10-20% overall unemployment within five years. Sam Altman thinks AGI will arrive during Trump's second term. Demis Hassabis says we'll have AGI in 5-10 years.
Meanwhile, experts give just a 23% chance of a "rapid progress" scenario by 2030—one where AI writes Pulitzer-worthy novels, collapses years of research into weeks, and independently develops cancer cures. The gap between what CEOs promise and what experts predict is absolutely massive.
And here's the strangest part: there's almost no disagreement among different expert groups. Computer scientists, economists, industry professionals, and policy experts all largely agree on these timelines. They just don't agree with their own CEOs.
Why this matters—and what we need to do about it:
Let's connect the dots. MIT shows 11.7% workforce displacement is technically possible today. Experts predict 18% of work will be AI-assisted by 2030. CEOs predict total job replacement by 2030.
Regardless of who's right, we're facing massive workforce transformation in the next five years. And here's the sobering reality: we're completely unprepared.
Let me put this in perspective like so. In The Changing World Order, Ray Dalio studied 500 years of empire rises and declines. His finding: education is the single most important factor in whether nations survive periods of transformation. Rising powers invest heavily in education and workforce development. Declining powers accumulate debt, cut investments, and fail to adapt when circumstances change (p.s: you can check out the great charts from the book here).
We're staring at 12% workforce displacement. Companies can't just squeeze workers out; they need retraining programs, and ones that actually work. Our societies can't keep running an education system where people spend 4-12 years learning skills for jobs that vanish before graduation. Policymakers can’t afford mass layoffs (like ~21M people) without equally massive social upheaval. So we gotta ask ourselves:
- What's the fastest way to retrain millions of workers?
- Which skills actually matter in five years?
- What industries need workers they can't find, and how do we align market incentives to get displaced workers there?
- Should certain industries get AI carve-outs to preserve employment?
- Will mass-unionization become the primary defense mechanism against wholesale automation?
- What other solutions exist?
Here's what we know for sure: whether it happens in 2 years or 10 years, the transformation is coming. Project Iceberg gives us the hard data, and the superforecasters give us a benchmark to plan towards. Now we need to decide if we're building the infrastructure and retraining systems to handle what's coming… or just watching the ‘berg drift closer while we debate whether there's room for Jack on the door with Rose.
