AI creates new tasks alongside automation
New research analysing 7.7 million German job vacancies and administrative data from 2017-2023 has found that AI didn’t just automate existing work—it created entirely new tasks, what economists call the reinstatement effect. This expansion of the task space toward analytic and interactive activities generated earnings up to 0.8% for workers exposed to these emerging skill requirements.
The study found AI exposure increased demand for analytical tasks by 4.7%—nearly one-and-a-half times the baseline average of analytic requirements per job advertisement. Interactive task spaces also expanded, whilst manual task categories contracted. This task creation represents an additional way through which workers benefit from AI, beyond simple productivity gains from existing work.
However, these benefits were distributed unevenly. Expert workers—those with deep domain knowledge—captured the strongest gains, with earnings increases of approximately 0.65% when AI skill demand doubled. Workers in cognitive-intensive roles with balanced task compositions, such as accounting and auditing, see particularly strong returns. In contrast, lesser-skilled workers earnings and employment stability declined.
Variation in AI skill demand across German local labour markets 2017-2023. Early AI exposure strongly predicted longer-term demand patterns, with markets displaying AI skills nearly doubling from 9% to 16%.
Overall, workers in the lowest earnings decile lost eight working days annually and experienced 3.9% earnings decline, whilst those in higher deciles gained up to five working days and 2.5% higher earnings. These patterns suggest AI may deepen existing labour market inequalities rather than level the playing field.
The findings highlight that job-specific expertise predicts AI-related outcomes more accurately than traditional education or experience measures. This underscores the growing importance of domain knowledge coupled with adaptable skills as AI adoption progresses across industries.