The number is staggering: $2.4 trillion in wages at risk from AI automation. But the real story isn’t the total — it’s which jobs are first in line, and why most of the damage is invisible until it’s already done.
MIT’s Project Iceberg, a research initiative combining the MIT Media Lab with Oak Ridge National Laboratory, has produced the most detailed skills-based analysis of AI’s labor market impact to date. Their Iceberg Index reveals that for every dollar of wage disruption visible in tech industry statistics, roughly five dollars of wage value is at risk in roles that don’t consider themselves technology-dependent.
And coding jobs are where the visible disruption begins.
The Iceberg Explained
The Iceberg Index uses an agent-based model simulating 151 million individual workers across 923 occupations in 3,000 US counties, decomposed into 32,000+ distinct skills. This granularity matters because AI doesn’t replace jobs — it replaces skills within jobs.
The key findings:
- Surface exposure (2.2% of wage value, ~$211 billion): Jobs where AI adoption is visible — software development, data science, and other explicitly tech-focused roles. These workers see disruption first and most visibly.
- Subsurface exposure (11.7% of wage value, ~$1.2 trillion): Roles where AI automates significant skill components but the job titles don’t sound “tech” — administrative services, financial analysis, legal document review, healthcare documentation, and professional coordination.
- Combined total exposure: Approximately $2.4 trillion in wages where AI can now replicate skill components that previously required human labor.
The name is deliberate. Like an iceberg, most of the mass is below the waterline. The jobs making headlines — tech layoffs, coding job elimination — are just the visible tip.
Why Coding Jobs Are First
The study identifies coding and software development as the earliest and deepest targets for AI displacement. This isn’t speculation — it’s already measurable:
- AI systems generate over a billion lines of code daily
- Companies are restructuring hiring pipelines, reducing demand for entry-level programmers
- Anthropic CEO Dario Amodei publicly stated that AI specifically targets programmer roles
- Q1 2026 saw 78,557 tech layoffs, with approximately 50% explicitly AI-attributed
The structural reason is that coding tasks are decomposable. A software engineer’s job bundles system architecture, code review, debugging, stakeholder communication, and technical writing alongside code generation. AI can automate code generation — the single largest time component — while leaving the judgment-heavy skills intact.
The result isn’t “no more programmers.” It’s fewer programmers doing higher-level work, with entry-level positions — the ones that involve mostly writing code — disappearing fastest.
The 5X Multiplier
The Iceberg Index’s most alarming finding is the 5X multiplier: subsurface exposure is approximately five times the visible surface exposure.
This means the current public debate — focused on tech layoffs and coding job losses — is dramatically underestimating the scope of disruption. For every software engineer who loses their job, approximately five other workers in administrative, financial, legal, and healthcare roles face significant skill-level automation that doesn’t show up in layoff statistics but steadily erodes their labor value.
The multiplier works because:
- Skills transfer across occupations — the same AI capability that writes code also writes legal briefs, financial reports, and medical documentation
- Subsurface roles have less awareness — a paralegal doesn’t see themselves as competing with AI, even as AI automates 60% of their core tasks
- Measurement tools are Industrial Age — GDP and employment statistics can’t detect skill-level displacement within intact job categories
What Makes This Study Different
Previous AI job impact studies relied on aggregate employment statistics or occupation-level analysis. The Iceberg Index’s methodological leap is skills-based decomposition combined with agent-based simulation.
Rather than asking “will AI replace accountants?”, the researchers ask “which of the 32,000+ skills that make up an accountant’s work can current AI systems replicate?” This reveals partial automation within roles — a more accurate and more concerning picture.
The agent-based model also captures non-linear dynamics: a 10% productivity improvement through AI doesn’t translate to 10% workforce reduction uniformly. It might mean 30% reduction in some roles, 2% in others, and creation of entirely new roles elsewhere, with timing varying by geography, industry, and firm size.
Second-Order Effects Nobody Is Modeling
The study acknowledges that even its $2.4T figure is likely conservative because it captures only direct wage exposure, not:
- Spatial multipliers: Each tech worker supports approximately 4.9 additional local jobs (Moretti multiplier). Tech layoffs ripple through food services, real estate, retail, and personal services.
- Capital reallocation: AI productivity gains accrue to shareholders and tech hubs. Job losses hit administrative centers and professional services markets. A laid-off paralegal in Cleveland doesn’t benefit from NVIDIA shareholder returns.
- Skills mismatch: New AI-created jobs aren’t fungible with destroyed ones. The economy needs prompt engineers, MLOps specialists, and AI ethicists — not administrative assistants and junior analysts.
What This Means for Workers
The research suggests several practical implications:
If you’re in coding or software development: You’re in the surface exposure zone. The disruption is visible and accelerating. The survival strategy isn’t to compete with AI on code generation — it’s to develop the judgment, architecture, and stakeholder skills that AI can’t replicate.
If you’re in administrative, financial, legal, or healthcare roles: You’re in the subsurface zone. The disruption is happening now but isn’t being measured. Audit your daily tasks: what percentage involves information processing, documentation, pattern recognition, or routine analysis? That percentage is your exposure.
If you’re in New Zealand: The Iceberg Index maps US counties, but the skill decomposition applies globally. NZ’s economy — heavy on professional services, agriculture, and tourism — has significant subsurface exposure. The absence of headline-grabbing tech layoffs doesn’t mean absence of AI wage erosion.
The Measurement Crisis
Perhaps the most important finding isn’t the $2.4T number — it’s that we don’t have the tools to measure what’s happening. Traditional productivity metrics, designed for factories producing widgets, collapse when applied to cognitive services. When an LLM helps a physician complete documentation in 15 minutes instead of 45, there’s no unit of output, no quality benchmark, and no price signal that GDP accounting can capture.
Policymakers are making trillion-dollar decisions with instruments that cannot detect the phenomenon they’re trying to regulate.
