AI Can Already Replace 11.7% of Jobs. Is Yours One of Them?

Nox Team·16 March 2026

In November 2025, researchers at MIT and Oak Ridge National Laboratory published a study that put a hard number on something most people had only guessed at: 11.7% of the U.S. workforce is already economically replaceable by current AI systems. Not future systems. Not theoretical breakthroughs. The technology that exists right now.

That figure translates to roughly $1.2 trillion in annual wages across 151 million workers. And Anthropic's Economic Index, released in early 2026, shows exactly which occupations are feeling it first.

The question is no longer whether AI will change the labor market. The question is whether your job is above or below the waterline.

The Iceberg Index: What Lies Beneath

MIT's study, called the Iceberg Index, draws a sharp distinction between what's visible and what's lurking underneath.

The visible part is familiar. Tech layoffs, shrinking engineering teams, AI coding assistants replacing junior developers. That slice accounts for about 2.2% of total U.S. wage exposure, or roughly $211 billion. It's concentrated in computing and information technology, mostly in coastal metro areas.

But beneath the surface sits a far larger mass. The remaining 9.5% of wage exposure spans human resources, logistics, finance, office administration, and professional services. It's geographically distributed across all 50 states, not confined to San Francisco and New York. It includes routine document processing, financial analysis, scheduling, and administrative coordination -- tasks that don't make headlines when they get automated because no one writes a press release about eliminating a back-office process.

The Iceberg Index is deliberately conservative. It doesn't measure what AI might theoretically do. It measures where AI can perform the same tasks at a cost that is competitive with or cheaper than human labor. These aren't projections. They're current economic realities waiting for adoption to catch up.

Anthropic's Data: What AI Is Actually Doing Right Now

If MIT's study maps the terrain, Anthropic's Economic Index provides the satellite imagery.

Based on privacy-preserving analysis of two million real AI conversations -- split between consumer use on Claude.ai and enterprise API usage -- the index measures observed exposure: the degree to which AI is already being used to perform occupational tasks in the real world.

The top 10 most-exposed occupations, ranked by observed exposure:

Rank	Occupation	Observed Exposure
1	Computer programmers	74.5%
2	Customer service representatives	70.1%
3	Data entry keyers	67.1%
4	Medical record specialists	66.7%
5	Market research analysts	65.0%
6	Sales representatives	63.0%
7	Financial and investment analysts	57.0%
8	Software QA analysts	52.0%
9	Information security analysts	49.0%
10	Computer user support specialists	47.0%

Three things stand out.

First, the top three are not surprises. Programming, customer service, and data entry have been cited as AI-vulnerable for years. What's new is the magnitude. Three-quarters of programming tasks are already being performed with AI assistance or automation. That's not a forecast -- it's a measurement of current behavior.

Second, the middle of the list is where white-collar professionals should pay attention. Market research, sales, financial analysis -- these are roles that require judgment, communication, and domain knowledge. They were supposed to be safe.

Third, the gap between theoretical and observed rates is the real story. The theoretical exposure for computer and math occupations is 94%. The observed rate is 33%. That 61-percentage-point gap represents the distance between what AI can do and what organizations have actually deployed. It's closing.

Theoretical vs. Observed: The Gap That Matters

This distinction is the most important nuance in the current data -- and the one most frequently lost in headlines.

Theoretical exposure measures whether an AI system could, in principle, perform a task at least twice as fast as a human. By this measure, the numbers are staggering. Goldman Sachs estimates that generative AI could affect 300 million jobs globally. McKinsey pegs the automation potential at 57% of all U.S. work hours. These figures describe capability, not deployment.

Observed exposure measures what's actually happening. And what's actually happening is more selective but accelerating in specific domains.

The St. Louis Federal Reserve published a study in August 2025 that found a 0.57 correlation between AI adoption intensity and unemployment rate increases by occupation. Computer and mathematical occupations -- the most AI-exposed category at roughly 80% theoretical exposure -- saw unemployment rise by 1.2 percentage points between 2022 and 2025. The researchers noted this is correlation, not proven causation. But the pattern is consistent across multiple data sources.

A Stanford Digital Economy Lab study sharpened the picture further: employment among software developers aged 22-25 declined nearly 20% from its late-2022 peak to mid-2025. Entry-level tech postings fell by as much as 67% in some categories. The learning curve -- the part of a junior developer's job that involves repetitive, well-documented tasks -- is precisely the part AI handles best.

Who's Actually Vulnerable?

Not everyone in an "exposed" occupation faces the same risk. A Brookings Institution study introduced the concept of adaptive capacity -- a composite measure of financial reserves, transferable skills, local job market density, and age.

Of the 37.1 million U.S. workers in the top quartile of AI exposure, 26.5 million also have above-median adaptive capacity. They have savings, marketable skills, and live in areas with diverse job markets. They'll adapt.

The remaining 6.1 million workers -- about 3.9% of the workforce -- face both high exposure and low adaptive capacity. They tend to be concentrated in clerical and administrative roles, disproportionately in smaller metro areas and college towns in the Mountain West and Midwest. 86% are women. These are the workers for whom "exposure" is most likely to translate into actual displacement.

The demographics of AI exposure run counter to the popular narrative. The Anthropic data shows that workers in the highest-risk professions tend to be older, more educated, and better-paid than average. AI isn't coming for factory floors first. It's coming for offices.

The Augmentation Question

The data isn't entirely grim. Anthropic's report found that 52% of AI usage on Claude.ai is augmentation -- humans using AI to do their existing jobs better -- versus 45% automation, where AI performs tasks independently. That ratio has been relatively stable, with augmentation consistently leading.

But the trend line matters. The share of jobs using AI for at least a quarter of their tasks has risen from 36% in January 2025 to 49% in pooled data through early 2026. The World Economic Forum's Future of Jobs Report projects a net gain of 78 million jobs by 2030 (170 million created, 92 million displaced). The jobs being created, however, require different skills than the ones being eliminated.

The fastest-growing roles are AI specialists, big data engineers, and fintech professionals. The fastest-declining are postal clerks, bank tellers, and data entry operators. The labor market isn't shrinking -- it's reshuffling. Whether that reshuffling works out for any given individual depends on their ability to move from one category to the other.

What the Data Actually Tells You to Do

The research converges on several practical conclusions.

Audit your task composition, not your job title. AI doesn't replace "jobs" in a single stroke. It replaces tasks. If 60% of your daily work involves things AI can already do -- formatting documents, writing first drafts, answering routine questions, processing structured data -- the economic pressure on your role is real and growing. The MIT Iceberg Index specifically measures task-level exposure, and that's the right unit of analysis for personal career planning.

The gap between theoretical and observed exposure is your window. The 61-point gap in computer and math occupations means adoption hasn't caught up to capability yet. That gap is time to learn the tools, reposition toward higher-judgment work, or move into roles where human skills remain essential. But it's a closing window, not a permanent one.

Geographic and demographic factors matter more than most people assume. If you're in a clerical role in a small metro area with limited job market diversity, your risk profile is materially different from someone in the same role in a major city. The Brookings adaptive capacity framework suggests that financial reserves and skill transferability are the two most actionable factors individuals can control.

Precision beats volume in a saturated market. With employers receiving 250+ applications per posting and only about 2.4% of applicants reaching the interview stage, submitting more generic applications isn't a strategy. The competitive advantage has shifted to application quality: tailored materials, targeted companies, and fast response to new postings.

The Bottom Line

The 11.7% figure from MIT is not a ceiling. It's a floor, measured with today's technology at today's adoption rates. Anthropic's data shows that observed exposure is rising quarter over quarter. The St. Louis Fed finds correlations between that exposure and actual unemployment increases. And the Brookings data identifies millions of workers who lack the adaptive capacity to navigate a transition.

None of this means the labor market is collapsing. The WEF projects net job growth through 2030. Augmentation still leads automation. Most exposed workers have the skills and resources to adapt.

But the data is clear: the workers who will fare best are the ones who understand their exposure, invest in transferable skills, and approach the job market with precision rather than volume.

Nox is an AI job application agent that scans 400,000+ job listings across 4 major ATS platforms and 13,000+ companies to find roles that match your skills, experience, and preferences. It writes tailored cover letters in your voice and submits applications on your behalf.

Try Nox free -- no credit card required.

Sources