Andrej Karpathy — co-founder of OpenAI and former head of AI at Tesla — published a ranking of 342 occupations scored by their exposure to AI automation. The analysis used U.S. Bureau of Labor Statistics occupation data and assigned each role an exposure score from 0 to 10 based on how easily AI could perform the job's core tasks.
The result was uncomfortable: the highest-exposure jobs aren't factory workers or truck drivers. They're the white-collar knowledge workers who thought they were safe — software developers, financial analysts, writers, editors, and graphic designers.
Key Takeaway
"Exposure" doesn't mean "replacement." It means AI can perform significant parts of the job. The workers who thrive will be those who use AI to amplify their judgment, not those who compete with AI on raw output speed.
Which Jobs Are Most Exposed to AI?
| Rank | Occupation | Exposure Score (/10) | Why |
|---|---|---|---|
| 1 | Computer Programmers | 9.5 | AI writes, debugs, and refactors code |
| 2 | Financial Analysts | 9.2 | AI analyzes data, builds models, writes reports |
| 3 | Writers/Editors | 9.0 | AI generates, edits, and formats text |
| 4 | Graphic Designers | 8.8 | AI generates images, layouts, and designs |
| 5 | Database Administrators | 8.7 | AI writes queries, manages schemas, optimizes |
| 6 | Mathematicians | 8.5 | AI solves equations, proves theorems |
| 7 | Accountants | 8.3 | AI processes transactions, detects anomalies |
| 8 | Market Research Analysts | 8.1 | AI surveys, analyzes sentiment, generates reports |
Which Jobs Are Least Exposed?
The pattern is clear: jobs requiring physical presence, human touch, or unpredictable physical environments are least affected.
| Occupation | Exposure Score (/10) | Why It's Safe |
|---|---|---|
| Construction workers | 1.2 | Physical, unpredictable environments |
| Janitors/Cleaners | 1.0 | Physical manipulation in varied spaces |
| Roofers | 0.8 | Dangerous physical work requiring judgment |
| Ironworkers | 0.7 | Skilled physical labor in hazardous conditions |
The irony Elon Musk pointed out: the highest-paid knowledge workers are most exposed, while the lowest-paid physical workers are safest. AI automates thinking before it automates doing.
---📬 Getting value from this? We publish weekly on AI's impact on work and careers. Get it in your inbox →
---What Does "Exposure" Actually Mean?
This is the critical distinction most coverage gets wrong. Exposure ≠ replacement. A software developer with a 9.5 exposure score doesn't mean 95% of developers will lose their jobs. It means AI can perform 95% of the raw tasks involved in software development — writing code, debugging, testing, documentation.
But the 5% that AI can't do — architecture decisions, understanding user needs, strategic tradeoffs, team coordination, knowing when the AI is wrong — is where all the value concentrates. As Karpathy himself argued at Sequoia AI Ascent 2026: "You can outsource thinking. You cannot outsource understanding."
The realistic outcome for high-exposure jobs: fewer people doing more work, with AI handling the volume and humans providing the judgment. A team of 10 developers becomes a team of 4 developers with AI agents. The 4 who remain are the ones who can direct AI effectively, catch its mistakes, and make architectural decisions.
How Do You AI-Proof Your Career?
1. Learn to use AI, not compete with it. The threat isn't AI replacing you — it's someone who uses AI replacing you. If you're a writer, learn to use Claude for first drafts and spend your time on strategy and editing. If you're a developer, learn Claude Code and focus on architecture.
2. Develop judgment that AI lacks. AI produces output. Judgment evaluates whether that output is right. Understanding your domain deeply enough to know when AI is wrong is the most valuable skill in any AI-exposed field.
3. Build AI-augmented workflows. Don't just use AI for individual tasks — redesign your entire workflow around AI capabilities. The professionals who thrive will be those with the best AI workflows, not the best raw skills. The ICCSSE prompting framework is a good starting point for structuring AI interactions.
4. Focus on what's unautomatable. Client relationships, stakeholder management, creative vision, ethical judgment, and leadership require human presence and trust. Build your career around these elements, using AI to handle the commodity work.
For practical tools to improve your AI workflow today, try the free Prompt Optimizer and browse our 49 free AI tools.
---📬 Want more like this? We write weekly on AI's impact on careers and work. Subscribe free →
---Frequently Asked Questions
Will AI actually replace software developers?
Not replace — reshape. AI will handle more implementation work (writing code, debugging, testing). Developers who can architect systems, evaluate AI output, and make strategic decisions will be more valuable than ever. Developers who only write boilerplate will struggle.
Is this analysis reliable?
Karpathy's methodology is sound — BLS occupation data scored against current AI capabilities. But exposure scores measure theoretical capability, not market reality. Regulatory, organizational, and trust barriers slow actual adoption. High exposure doesn't mean immediate disruption.
How long until these impacts are felt?
Already happening in tech. Layoffs at major companies have been partially attributed to AI productivity gains requiring fewer workers. Other industries will follow over 2-5 years, with regulated industries (finance, healthcare, legal) moving slower due to compliance requirements.
Should I switch to a "safe" career?
Switching to construction to avoid AI is not the answer. The better strategy: stay in your field but become the person who uses AI most effectively. Every high-exposure field will still need humans — just fewer of them, and the ones who remain will be those who combined domain expertise with AI fluency.
Disclosure: Some links in this article are affiliate links. We only recommend tools we've personally tested and use regularly. See our full disclosure policy.