As generative artificial intelligence (AI) moves rapidly from promise to practice, its early labor-market effects pose a puzzle for economists. Standard frameworks of skill-biased and routine-biased technological change predict that new technologies either complement highly skilled workers or displace routine labor. Yet emerging evidence paints a more complex picture. Recent results from Brynjolfsson et al. (2025a) show that the initial impact of widespread AI adoption is characterized by a pronounced contraction in employment-rather than wages-concentrated among highly educated, entry-level workers in prediction-intensive occupations, with little discernible effect on aggregate employment. We argue that these patterns cannot be fully understood within existing paradigms and call for a new conceptual synthesis. By integrating task-based macroeconomic models of automation with a microeconomic view of AI as a technology that substitutes for implementation skills while complementing human judgment, we delineate the backbone of a unified framework to analyze how AI is likely to reshape job design, reallocate comparative advantage across tasks, and alter the distributional consequences of technological change. This perspective serves to structure a formal research agenda on the evolving interaction between AI and labor markets and its implications for productivity, employment, and inequality. 

This paper develops a model in which strategic complementarities emerge in a banking system constrained by supervisory frictions. The supervisory authority operates under a binding budget constraint that limits its monitoring capacity, while compliance with regulatory oversight imposes costs on banks. Within this setting, macroprudential policy seeks to regulate aggregate credit dynamics by preventing the system from converging toward extreme equilibria - either excessive credit expansion or a credit freeze. We show that achieving this objective requires a two-pronged policy approach: countercyclical capital buffers effectively curb excessive credit growth during booms, while loan-support programs sustain credit provision in downturns. No single instrument is sufficient to ensure financial stability; instead, a combination of regulatory tools is required to address distinct phases of the financial cycle and counteract the amplification mechanisms driving systemic risk.

Accurate and timely statistics on business dynamics are essential for analyzing firm turnover, job creation, and sectoral contributions to aggregate activity. A key challenge in this context is distinguishing genuine new entrants from reactivated firms—entities that re-emerge after apparent exit but retain strong continuity with previously closed businesses. Official data sources often conflate these categories, leading to biased estimates of entry and exit rates. This paper proposes a NLP-based methodology to improve the classification of firm transitions by combining high-frequency administrative records with unstructured textual data. We first use supervised learning applied to firms’ self-declared business descriptions to impute missing sectoral classifications and introduce a filtering algorithm to isolate economically informative text. Secondly, we estimate a Random Forest model to predict firm reactivation and distinguish true entrants from regenerated firms. The resulting framework enhances the quality and timeliness of firm-level statistics and offers new tools for studying business demography and its micro- and macroeconomic implications.