What does Ricardo's theory of comparative advantage tell us about a future where AI is better than Humans at every task? To find out, I started experimenting with tools from trade theory, which grew into a working paper (link below). The answer is more hopeful than you might expect. And Ricardian assignment models turn out to be well suited to measuring the effects of *jagged* AI growth, whether the jaggedness is in intelligence or in adoption. Key takeaways: (1) Data reveal that AI's capability frontier is more jagged than past labor-saving technologies like trade. However, agentic AI has more breadth, and may outperform Humans in virtually all tasks. (2) If AI capabilities keep growing at this pace but stay jagged, average wages rise, even as AI pushes workers out of large swaths of tasks. (3) At current adoption levels, every occupation could see wage growth, even if workers can't easily switch sectors. With more jagged adoption, some would risk immiseration through Baumol's cost disease. (4) Despite its scale, AI growth may be less immiserizing than past labor-saving technologies like trade or engines, because Its capabilities are more jagged, and its adoption, while still low, is more even across sectors. A thread 🧵