No more forecasts – real data
For years, economists, management consultants and technologists have been predicting which professions will disappear due to artificial intelligence. The predictions were mostly dramatic – and just as often wrong. The panic about outsourcing in the early 2000s cost hardly any jobs as forecast. The wave of automation in industry proceeded more slowly and selectively than expected.The fundamental problem with these forecasts has always been the same: they were based on what machines could theoretically achieve – not on what they actually do on a daily basis. A language model can theoretically formulate legal documents, produce financial reports, and debug code. But how often does this actually happen? To what extent? In which professions?
A study by researchers at Anthropic now answers precisely this question: Maxim Masenkov and Peter Macroory have developed a new metric – the so-called Observed Task Coverage (OTC). And the results are as astonishing as they are revealing.
The Method: Theory Meets Reality
The researchers combined three data sources:- O*NET – the US state occupational classifier. This database breaks down every occupation into precise micro-tasks with time allocations.
- Theoretical AI performance evaluation – assessments of how much AI can accelerate tasks.
- Real-world usage data of the Claude model – what tasks people actually delegate to AI.
The Food Processor Paradox
To illustrate the core problem of earlier studies, the researchers use a vivid analogy:A food processor can theoretically cook a five-course meal – in practice, it's mostly used for chopping onions.
The same applies to AI. The central question isn't what it can do – but what is actually being used. The answer: we're still in the "onion-chopping" phase.
The figures: The gap between potential and reality
An example from practice:- Theoretical AI potential (IT professions): 94 % of tasks
- Actual usage: only about 3 %
- Legal liability (e.g. medicine, law)
- Outdated IT infrastructure in companies
- Human control and decision-making processes
The limitation is not technological – but institutional.
Who is truly at risk? The ranking
| Rang | Field of work | Observation Cover | Main task |
|---|---|---|---|
| 1 | Programmer | 74,5 % | Code schreiben und pflegen |
| 2 | Customer support | 70 % | Communication |
| 3 | Data entry | 67 % | Data processing |
| 4 | Medical Documentation | ~55 % | Create reports |
| 5 | Marketing analysis | ~50 % | Reporting |
| 6 | Sales | ~45 % | Offers & Follow-ups |
| 7 | Financial analysis | ~44 % | Forecasts |
| 8 | Software testing | ~40 % | Test cases |
| 9 | IT Security | ~38 % | Threat analysis |
The Paradox of Education
A surprising result:- 17 % from the at-risk group have a master's degree
- Only 4.5 % of this group are barely affected by AI
The higher the qualification, the higher the AI exposure, often.
Tasks that are purely cognitive – writing, analysing, structuring – are particularly well-suited for automation.
Where are the mass layoffs?
Die Antwort: Sie passieren indirekt. Unternehmen entlassen keine erfahrenen Mitarbeiter. Stattdessen: The answer: They happen indirectly. Companies do not lay off experienced employees. Instead:- Senior staff become more productive
- Fewer junior positions are being created
- New hires (aged 22–25) in AI roles: −14% % since 2022
The door to the job market is quietly closing – not visibly.
Long-term effects
Statistical models show:- +10 % AI coverage → −0.6 % employment growth
Strategies: What to do?
For experienced professionals
- Using AI as a productivity booster
- Automate workflows
- Take on strategic tasks
For those starting their careers
- Focus on skills that AI cannot replace
- Communication, judgment, responsibility
For non-academics
- Craftsmanship and physical labour remain stable
- High future security in practical professions
For career changers
- Choosing professions with social interaction
- Mastering AI as a tool
Conclusion
The Anthropic study doesn't show an extreme scenario – but rather a nuanced reality. AI won't suddenly replace millions of jobs. Instead, it will change them:- Access to the labour market
- Productivity requirements
- Value creation within professions
The crucial question isn't: "Will I lose my job?" – but rather: "Am I using AI better than others?"
The revolution is happening – quietly, but profoundly.
