Using more than 120 million online job postings in France since 2019, we document the diffusion of artificial intelligence across firms, sectors and regions, and study its effects on labor demand. We find that AI adoption is highly concentrated and is associated with higher productivity, revenues and employment, together with a reallocation of hiring across occupations.

Using weekly model-level usage and price data from OpenRouter, combined with capability measures from Artificial Analysis, this project examines the creation and destruction of innovation rents in the AI-model market. Frontier releases generate large but short-lived gains, while entry by cheaper comparable-quality models erodes incumbent usage and market share through substantial business stealing. Yet leading providers continue to benefit from rapid market expansion and retain a segment of users willing to pay for premium-quality models. These dynamics are rationalized with a Schumpeterian model of frontier innovation and low-cost catch-up.

We introduce an internationally consistent definition of metropolitan areas for remote-sensing applications and apply it to 687 European metropolitan areas. Combining Sentinel satellite imagery with deep learning, we show that the choice of urban boundaries matters for both estimated land use and classification accuracy.

We develop a deep-learning approach to monitor urban land use from Sentinel satellite imagery. Applying a U-Net model to 687 European metropolitan areas, we map residential and business-related land use and provide a scalable framework for tracking urban development consistently and in near real time.