How Exposed Is Your Job to AI? A Top AI Researcher Scored Every US Occupation

You just Googled your job title and "AI." We know. Half the internet did the same thing this week, because one of the most credible people in artificial intelligence just scored every occupation in America.

Here is what the data actually says, what it misses, and what you can do about it.

What Karpathy Built

On March 14, 2026, Andrej Karpathy published an open dataset on GitHub scoring all 342 Bureau of Labor Statistics occupations for AI exposure. If you do not know the name: Karpathy was the founding member of OpenAI and former director of AI at Tesla. He is not a pundit. He builds the systems everyone else writes think-pieces about.

Update: The GitHub repository has since been removed. The data and methodology described below were verified while the repo was live. We are preserving this coverage because the framework and findings remain relevant, even without the original source link.

His method was straightforward. He fed every BLS occupation description into Google's Gemini Flash model and asked it to rate AI exposure on a 0-to-10 scale. Zero means AI has almost no relevance to the work. Ten means the job could, as of March 2026, be handled almost entirely by AI.

The "Laptop Test"

Karpathy's core finding boils down to one sentence:

"If you can do your entire job from a home office on a computer, your AI exposure is 7 or higher."

That is his "laptop test." It is blunt. It is imperfect. And it is directionally hard to argue with.

The logic: AI systems, as of March 2026, live inside computers. They read text, write text, generate images, analyze data, and write code. If your job happens inside the same box the AI lives in, the overlap is enormous. If your job requires you to physically show up, touch things, or navigate unpredictable environments, AI has far less purchase.

This does not mean laptop workers lose their jobs tomorrow. It means their tasks are in AI's crosshairs first.

The Score Breakdown

Here is how 342 occupations distribute across Karpathy's 0-10 scale:

Score Range	Exposure Level	% of Occupations	Example Occupations
0-2	Minimal	~15%	Construction laborers (1), athletes (1), drywall installers (1)
3-4	Low	~18%	Electricians, plumbers, firefighters
5-6	Moderate	~29%	Teachers, nurses, marketing managers
7-8	High	~30%	Software developers (8), graphic designers (8), paralegals (9)
9-10	Very High / Maximum	~8%	Accountants (8-9), translators (9), medical transcriptionists (10)

Key stats: The average score across all 342 occupations is 5.3 out of 10. The mode, the most common score, is 7, shared by 70 occupations. Only one occupation scored a perfect 10: medical transcriptionists.

Roughly 38% of all occupations score 7 or above, placing them in the high-to-maximum AI exposure range.

What Your Job Scored

Here are scores for occupations people search for most on this site:

High exposure (7-9):

Software developers: 8 -- Code generation, debugging, and documentation are core AI capabilities as of March 2026.
Graphic designers: 8 -- Image generation tools have transformed production workflows.
Accountants and auditors: 8-9 -- Tax preparation, reconciliation, and compliance checks are heavily digitized.
Paralegals and legal assistants: 9 -- Document review, case research, and contract analysis map closely to what large language models do.
Writers and authors: 8 -- Drafting, summarizing, and editing text is AI's wheelhouse.
Financial analysts: 8 -- Modeling, forecasting, and report generation are increasingly automated.

Moderate exposure (4-6):

Teachers: 5 -- Lesson planning and grading are exposed, but classroom management and mentorship are not.
Nurses: 4 -- Documentation is exposed, but bedside care, patient assessment, and physical intervention are not.
Marketing managers: 6 -- Content creation and analytics are exposed, but strategic judgment and relationship management lag behind.

Low exposure (1-3):

Construction workers: 1 -- Physical, unpredictable environments remain largely beyond current AI and robotics.
Athletes: 1 -- Physical performance cannot be delegated to software.
Electricians: 2 -- Hands-on work in unstructured environments keeps exposure low.

The Five Caveats You Need to Know

Before you panic or celebrate, this dataset has real limitations. Karpathy himself flags several:

1. Single-model scoring. The ratings come from one AI model (Gemini Flash) in a single pass. A different model, or even the same model on a different day, might produce slightly different scores. There is no averaging, no human review panel, and no inter-rater reliability check.

2. Whole-occupation, not task-level. A software engineer scored 8, but that collapses a huge range of tasks. Writing boilerplate code (very exposed) and designing system architecture for novel problems (much less exposed) get the same number. Task-level analysis, like the O*NET-based studies from Anthropic and OpenAI, paints a more granular picture.

3. Digital-work bias. The "laptop test" insight is powerful, but it also means the scoring framework tilts toward measuring digital task overlap. It may underweight the non-digital components of otherwise digital jobs. A lawyer's courtroom presence, for instance, matters but may not show up in a BLS job description.

4. US-only. The BLS taxonomy covers US occupations. Labor markets, regulation, and adoption curves differ globally. A score of 8 for software developers in the US may not translate directly to markets with different labor dynamics.

5. Static snapshot. These scores represent AI capabilities as of early 2026. AI is as bad as it will ever be right now. A job scoring 4 today could score 7 in eighteen months if robotics or multimodal AI leaps forward. Conversely, regulatory changes could slow adoption in high-scoring fields.

Reality check: This is a useful directional tool, not a career death sentence. A score of 8 means high exposure to AI capability overlap. It does not mean 80% of people in that role lose their jobs next quarter.

What to Do About Your Score

Whatever number your occupation landed on, here is how to think about next steps.

If you scored 7+: Your job's tasks overlap heavily with what AI can do today. That is not the same as being replaced. It means the way you do your job is changing fast.

Audit your daily tasks. Which ones could AI handle at 80% quality? Those are the ones your employer will eventually automate or expect you to accelerate with AI tools.
Identify your judgment layer. The tasks where context, relationships, and accountability matter most are your moat. Lean into them.
Learn the tools. The professionals who thrive at score-8 jobs are the ones using AI as a multiplier, not competing with it on raw output.

If you scored 4-6: You have time, but not unlimited time.

Watch the task edges. AI may not threaten your core role yet, but it is chipping away at the digital portions. Stay fluent with the tools entering your field.
Document your non-digital value. What do you provide that cannot be captured in a text prompt? Make that explicit to your employer.

If you scored 1-3: Your physical, unpredictable, or deeply human work is currently outside AI's reach. Do not get complacent.

Keep an eye on robotics. Humanoid robots from Tesla, Figure AI, and others are advancing. A score of 1 today does not guarantee a score of 1 in 2030.
Build cross-skills. The safest position long-term is combining physical expertise with digital fluency.

The Bottom Line

Karpathy's dataset is the most accessible occupation-level AI exposure map published in 2026 so far. It confirms what many suspected: the more digital your job, the more exposed you are. But exposure is not elimination. It is a signal to adapt, not a pink slip.

The professionals who come out ahead will be the ones who see a score of 8 and ask, "How do I use this?" instead of "Am I done?"

Your move: Spend 30 minutes listing the tasks in your day that AI could realistically handle. Then ask yourself: what do I do that AI cannot? That gap is your career strategy for the next three years.

Sources & Further Reading

Karpathy, A. (2026). "AI Exposure Scores for 342 BLS Occupations." GitHub repository. Published March 14, 2026. Repository has since been removed. Data was verified while live.

Related on this site:

Last Updated: March 2026