The Paper That Changed the Conversation
A medical transcriptionist in Phoenix woke up on March 15, 2026 to find Andrej Karpathy had published a 342-occupation analysis of AI job risk. Her job scored 10 out of 10. Not 7. Not 8. Ten. The ceiling.
Her specialty, converting physician voice recordings into structured clinical notes, is now done faster, cheaper, and at scale by AI. The BLS already shows an 8% job decline in that role. Karpathy just put a number on what she already felt.
That's the difference between vague anxiety and actionable data. And that data is what we need to talk about.
The scale of the shift
42% of US jobs now score 7 or higher on AI exposure. That's 59.9 million jobs and $3.7 trillion in wages in active restructuring territory.
Your Degree Is Not the Shield You Think It Is
Here's what most people get wrong. They assume AI targets the bottom of the labor market. Low-skill, low-wage, replaceable work. They're wrong.
Bachelor's degree holders average 6.7 on AI exposure. Workers without a degree average 4.1. The more credentials you carry, the more your work tends to involve language, analysis, and pattern recognition. Which is exactly what large language models do best.
Jobs paying over $100,000 average 6.7 on the Karpathy AI job risk scale. Jobs under $35,000 average 3.4. The plumber wins this one. The consultant doesn't.
Education got you into this bracket. It won't automatically protect you inside it. The credential is not the moat. The craft is.
42% of Gen Z is now actively pursuing trades. Plumbers and HVAC technicians score 0-2. Not because the work is easy. Because it requires embodied skill, physical presence, and contextual judgment that no model can replicate from a server rack.
That's not a consolation prize. That's a strategic insight.
High Score. Booming Demand. Understand the Difference.
Software developers score 8-9 on the 342 occupations AI analysis. By the simple logic most people apply, that means they're in trouble. Except job growth in that field is projected at +25%.
This is where Karpathy's framework earns its complexity. A high AI exposure score doesn't mean elimination. It means restructuring. It means the nature of the work is changing fast, the toolset is changing fast, and the people who adapt own the upside.
The adaptation premium
AI skills command a 56% salary premium right now. 81% of physicians use AI daily, up from 38% in 2023. Adaptation is not optional. It's already happening.
The actual danger zone is narrow. Only 3% of jobs score 9-10, near-full automation territory. Medical transcriptionists are there. Data entry clerks are there. Telemarketers are there. These aren't jobs being restructured. They're jobs being ended.
The bulk of exposure, the 7-8 range, means your role changes in 2-3 years. Not eliminated. Restructured. The question is whether you're doing the restructuring, or it's being done to you.
-
Score 9-10: Disruption happening now. These roles have already tipped. Medical transcriptionists, data entry, routine claims processing.
-
Score 7-8: Active restructuring window. 2-3 years to adapt or be left behind. The largest bucket.
-
Score 5-6: Meaningful exposure but 5+ year runway. Time to observe, learn, and build leverage before the pressure arrives.
-
Score 0-2: Physical presence and embodied judgment. Nurses, electricians, plumbers, physical therapists. The current floor of AI exposure.
Same Hospital, Completely Different Futures
Walk into any major hospital. You'll find surgeons, radiologists, nurses, and medical transcriptionists working in the same building, billing the same system, wearing the same scrubs.
Their AI exposure scores span the entire range.
Surgeons score 3. The physical manipulation of tissue, the real-time judgment calls, the haptic feedback loop, that's irreplaceable for now. Nurses score 2. Patient care, emotional attunement, physical presence are the job. Radiologists score 7. Reading imaging studies is pattern recognition at scale, which is exactly what AI does exceptionally well. Medical transcriptionists score 10.
Same employer. Same building. Radically different futures. The job title doesn't tell you where you stand. The tasks inside it do.
Your job title is a label. Your actual work is a portfolio of tasks. AI is coming for the tasks, not the label.
This reframe is the core insight from Karpathy's 342 occupations analysis. Stop asking "Is my job safe?" Start asking "Which of my daily tasks can be automated, and what's left when they are?"
What's left is your new job description. Build toward it now.
The Second-Order Effect Nobody's Talking About
Here's where it gets uncomfortable for managers and executives who feel safely insulated from this.
A VP of Sales scores 6 on AI exposure. Moderate. Not immediate. Probably fine for a few years. But the SDRs under them score 8. The BDRs score 8. The people doing outbound prospecting, initial qualification, and pipeline generation are being automated out or radically augmented right now.
When the bottom of a team structure gets automated, the top doesn't stay the same. The VP now manages AI tools and a smaller headcount. The role changes. The required skill set changes. The score of 6 is hiding a structural transformation that arrives faster because of what's happening at 8.
This is the second-order effect the 342 occupations AI analysis points toward. Your exposure isn't just about your role. It's about the roles your role depends on. If your entire pipeline feeds from people who score 7+, you're closer to disruption than your personal score suggests.
Where do you actually stand?
500+ occupations scored 0-10. Free. Takes 60 seconds.
What to Do With Your Score Right Now
A number between 0 and 10 is only useful if you act on it. Here's the framework.
Map your actual tasks, not your title. Write down the five things you do most in a given week. Not your job description. The actual work. Then score each task's AI replaceability independently. That composite is your real exposure.
Identify the tasks that remain after automation. What's left when AI handles the repeatable parts? That residual is where your value concentrates. Invest in getting exceptional at it. The 56% salary premium goes to people who master AI-augmented versions of these tasks.
Check the second-order exposure of your team. If the people who feed your work score 7+, your runway is shorter than your personal score implies. Plan accordingly. The restructuring arrives at the team level before it arrives at the title level.
Use your score's timeline as a planning horizon. Score 9-10: act now, the window is closing. Score 7-8: 2-3 years to build the augmented version of your role. Score 5-6: 5+ years but don't sleepwalk through them. Score under 4: you have strategic time, not infinite comfort.
Bottom Line
Karpathy's analysis of 342 occupations isn't a doomsday report. It's a diagnostic. The global average exposure sits at 5.3 out of 10. Most people have a window. Most people have choices.
The ones who will lose are not those with high scores. They're those who mistake their score for a verdict instead of a signal.
The data is available. The patterns are clear. The question is whether you'll use your timeline to adapt, or spend it assuming the disruption will stop before it reaches you.
It won't stop. But it will reward the people who moved early.
The future of work isn't being decided by AI. It's being decided by the people who understand AI well enough to stay ahead of it.
Find out where you stand
500+ occupations scored 0-10 on AI displacement risk. Free.
Keep Reading
Methodology: AI Displacement Scores are calculated using the JobHunter AI Displacement Index, which analyzes 500+ occupations across 12 risk factors including task automation potential, historical automation patterns, AI capability trajectories, and labor market dynamics. Data sources include Stanford's AI Index Report, Anthropic's capability research, Bureau of Labor Statistics employment projections, and O*NET task databases. Scores are updated quarterly. Learn more about our methodology.
Related AI Displacement Scores: Highest Risk · Lowest Risk · Tech Roles