What are the consequences of multidimensional skill uncertainty for workers’ wages and aggregate output? I develop and estimate a general equilibrium dynamic Roy model in which workers have imperfect information about their multidimensional skills and accumulate task-specific human capital. Estimated on Portuguese administrative data, the model rationalizes key patterns of occupational mobility, with learning about comparative advantage playing a central role in occupational reallocation among young and poorly matched workers. Removing information frictions raises aggregate output by 5.2%, primarily through better skill allocation across occupations, while generating the largest wage gains early in workers’ careers and among the most mismatched. A feasible information treatment about comparative advantage across tasks recovers about 18% of this output loss. Finally, job transformation induced by Large Language Model adoption raises output while widening the output gap between the imperfect- and full-information economies along the transition path and in the new long run.