4.5 million jobs. 71 occupational groups. One question: How much is generative AI changing the Austrian job market? The debate surrounding this is passionate – but rarely data-driven. The Austrian Occupation Exposure Map makes this question interactive and explorable: using official employment data, reproducible methodology, and open-source code.
- What the Austrian Occupation Exposure Map shows and how it works
- Where the data comes from and why it is reproducible
- What insights the visualisation provides
- How the open-source project is technically structured
- What companies, educational institutions, and policymakers can deduce from it
Table of Contents
What the Map Shows
The Austrian Occupation Exposure Map is an interactive treemap visualisation of the Austrian job market. Each tile represents an ISCO-08 occupational group. The area corresponds to the number of employees – the larger the tile, the more people work in this professional field.
Four Layers, Four Perspectives
The map offers four interchangeable colour layers (layers) that reveal different dimensions of the job market:
Outlook
Employment trend: Green signals growth, red indicates declining employment. Based on Eurostat time series.
Median Income
Gross annual salaries in EUR – including 13th and 14th month salaries, as is customary in Austria. Source: Structure of Earnings Survey 2022.
Education Level
Typical qualification level per occupational group – from compulsory schooling to university degree.
AI Exposure
How strongly generative AI is expected to change the respective occupational group. LLM scoring on a scale of 0 to 10.
In addition, the application offers five navigation views: Treemap, Job Explorer, ISCO families, sectors (ÖNACE), and Context & Method – a comprehensive methodology section with full source references, reproducible data processing documentation, and a dedicated sources page with direct download links to all primary datasets.
The Austrian Job Market in Figures
- Total employees
- 4.5 million
- Average outlook (job-weighted)
- +0.8%
- ISCO-08 occupational groups
- 71
| Outlook category | Employees | Share (%) |
|---|---|---|
| Declining (<0%) | 1.3 million | 30 |
| Slow (0–3%) | 2.6 million | 59 |
| Average (4–7%) | 368,000 | 8 |
| Fast (8–14%) | 129,000 | 3 |
A key finding: 30% of employees (1.3 million) work in occupational groups with a declining employment trend. The majority (59%) falls into the category of slow growth. Strong growth (over 8%) affects only 3% of employees. In total, 1.3 million employees in shrinking occupations face 2.7 million in growing ones.
Outlook by Earnings Bracket
The map additionally breaks down the employment outlook by income bracket – a revealing picture:
| Earnings bracket | Outlook |
|---|---|
| <€25K | +1.0% |
| €25–35K | −0.0% |
| €35–50K | +0.1% |
| €50–70K | +2.2% |
| €70K+ | +1.0% |
The strongest employment dynamics appear in the €50–70K bracket at +2.2%. The €25–35K range effectively stagnates at −0.0%. Neither the lowest nor the highest income brackets show the strongest momentum – it is the upper-middle tier that grows most clearly.
A high AI exposure score does not mean that professions will disappear. Software development, for example, reaches scores of 8–9/10 – but the demand for software increases when every developer becomes more productive. The score measures transformation, not elimination. Elasticity of demand, regulatory hurdles, and societal preferences are not priced in.
From Karpathy's Idea to the Austrian Adaptation
The US Original
In March 2026, Andrej Karpathy – former Director of AI at Tesla and co-founder of OpenAI – published the US Job Market Visualizer. The tool visualises 342 professions from the Bureau of Labor Statistics Occupational Outlook Handbook and covers 143 million jobs in the US economy.
What makes Karpathy's approach special: The AI Exposure Scores are not assigned manually, but generated via an LLM pipeline. A prompt defines the evaluation grid, and the model evaluates each job description individually. The entire process is reproducible and publicly accessible in the GitHub repository.
"This is not a report, a paper, or a serious economic publication — it is a development tool for exploring BLS data visually."
— Andrej Karpathy, US Job Market Visualizer
Karpathy's scoring prompt defines clear anchor points for the 0–10 scale:
| Tier | Criterion | Examples |
|---|---|---|
| Minimal (0–1) | Physical work, barely any AI influence | Roofers, landscaping |
| Low (2–3) | Predominantly physical/interpersonal | Electricians, plumbers, fire service |
| Moderate (4–5) | Mix of physical and knowledge work | Nursing staff, police, veterinarians |
| High (6–7) | Predominantly knowledge work | Teachers, management, accounting |
| Very high (8–9) | Almost entirely digital | Software development, graphic design, translation |
| Maximum (10) | Routine information processing | Data entry, telemarketing |
A key signal: Whether the work is fundamentally digital. If a job can be done entirely from a home office – writing, programming, analysing, communicating – the AI exposure is inherently 7+. Professions requiring physical presence, manual dexterity, or direct human interaction in the physical world have a natural barrier.
What Differs
The Austrian Occupation Exposure Map adopts Karpathy's concept and scoring rubric – but the data sources, information architecture, and methodology were completely rebuilt for the Austrian job market.
| Feature | US Original (Karpathy) | Austrian Adaptation (webconsulting) |
|---|---|---|
| Data sources | Bureau of Labor Statistics (BLS) | Eurostat lfsa_egai2d, nama_10_a64_e, Statistics Austria OGD |
| Occupational classification | SOC (342 professions) | ISCO-08 (71 occupational groups) |
| Structure | Sector-based | Occupation-based + ÖNACE sector context |
| Income | USD Median Annual | EUR gross incl. 13th/14th salary |
| Language | English | German / English (bilingual) |
| Route structure | Sector-based | /occupation, /family, /sector |
Methodology: Transparency Instead of a Black Box
LLM Scoring: How AI Exposure Is Evaluated
The scoring follows Karpathy's methodology: Each occupational group receives an AI exposure score on a scale from 0 to 10. The values are curated and defined in scripts/generate-occupations.ts – each with an English-language rationale. Optionally, scores can be automatically re-evaluated via an LLM using npm run score:exposure-llm; the generator merges the results into the dataset. The prompt defines clear evaluation criteria and anchor points following Karpathy's rubric (cognitive/digital task content).
LLM scoring pipeline: From job description to visualisation
The crucial point: The framework is extensible. The same pipeline approach works for any given question – exposure to humanoid robotics, offshoring risk, climate impact. A new prompt is all it takes, and the treemap is coloured according to the new criteria.
Data Sources and Reproducibility
Every data point can be traced back to an official source. All raw data can be freely downloaded as Open Government Data. The generation pipeline (scripts/generate-occupations.ts) is open-source, and 30+ automated integrity checks verify the dataset with every regeneration.
The data sources in detail:
- Employment: Eurostat
lfsa_egai2d(Microcensus Labour Force Survey 2024) by ISCO-08. Additionally,nama_10_a64_e(NACE sector employment) serves as proportional weights to split ISCO totals across the ÖNACE-labelled occupation rows. - Income: Structure of Earnings Survey 2022 (Statistics Austria Open Government Data) – two datasets:
OGD_veste403provides gross hourly median pay by ISCO-08 occupational group,OGD_veste401by economic section as supplementary sector context. Annual gross salary is calculated as: hourly median × 2,080 hours/year × 1.17 (13th and 14th month salary). - Sector context: ÖNACE 2025 (effective since 1 Jan 2025) for economic sections A–S. The VSE 2022 was conducted under ÖNACE 2008; at the section level the structure is identical.
- AI Exposure: Integer 0–10 per occupational group, curated in
scripts/generate-occupations.tsfollowing Karpathy's rubric (cognitive/digital task content). Optionally extensible vianpm run score:exposure-llmwith LLM-based re-evaluation.
What You Can Do with the Map
The Austrian Occupation Exposure Map is more than just a pretty graphic. It is an analytical tool for anyone who wants to think about AI and work in a data-driven way.
For Companies
Strategic workforce planning
Which occupational groups in your company have the highest AI exposure? Where is upskilling worthwhile, and where are new roles needed? The map provides the data foundation for evidence-based decisions.
Validating the technology roadmap
Compare your planned AI investments with the exposure scores. Where do internal assumptions align with the data – and where are there blind spots?
For Education and Policy
Developing curricula
Educational institutions can see at a glance which professional fields are changing the most. The education level layer shows where qualification requirements are increasing.
Evidence-based labour market policy
A combination of outlook and AI exposure: Which occupational groups are losing employment and have high AI exposure? This is where the most urgent need for action arises.
For Research and Journalism
- Reproducible analysis: All data and methods are open – ideal for scientific papers and investigative research.
- Custom scoring: The LLM pipeline can be extended with your own prompts. How exposed are Austrian professions to humanoid robotics? Write a prompt, run the pipeline.
- Cross-country comparison: The US original and several international adaptations (Australia, EU) enable direct comparisons.
Technology Under the Hood
The Tech Stack
- Next.js with App Router and Server Components
- TypeScript for end-to-end type safety
- shadcn/ui as a component library
- webconsulting Design System for consistent branding
- Responsive design with a bilingual interface (DE/EN)
Open Source and Reproducibility
The entire project is open-source and available on GitHub:
Austrian Occupation Exposure Map – GitHub Repository
Complete source code of the Austrian Occupation Exposure Map: data pipeline, LLM scoring, visualisation, and 30+ integrity checks. Fork it, extend it, build on it.
Andrej Karpathy's US Job Market Visualizer – GitHub Repository
The original project: 342 US professions, LLM scoring pipeline, and interactive treemap visualisation. The template for all international adaptations.
What Could Be Possible Next
The open architecture invites further development. Some approaches that could build on the existing framework:
Conclusion
The Austrian Occupation Exposure Map makes an abstract debate tangible. Instead of speculating about "AI and work", companies, educational institutions, and policymakers can concretely analyse 71 occupational groups and 4.5 million jobs – with official data, transparent methodology, and open-source code.
It is not the technology that decides which professions will change – but rather how quickly organisations recognise and shape this change.
The key takeaways:
- Data-driven instead of speculative: All employment and income data come from official sources (Eurostat, Statistics Austria OGD) and are fully reproducible.
- Transformation, not elimination: High AI exposure means a change in working methods – not necessarily fewer jobs. Elasticity of demand and societal factors play a central role.
- Extensible by design: The LLM scoring pipeline can be applied to any question using new prompts.
- Open-source: Source code, data pipeline, and integrity checks are publicly available on GitHub – to fork, extend, and further develop.