This paper uses artificial intelligence-enhanced agent computing to determine how to allocate budgetary resources within a large set of heterogeneous government programs targeting human capital. The approach considers essential features of the budget allocation process: multidimensionality, interdependencies between policy issues, and the political economy of public officials‘ collective action. The analysis uses highly disaggregated Mexican data covering the 2016-2022 period across 49 federal government human capital programs. It focuses on how expenditure affects program coverage, defined as the share of the population with a public problem and that has access to various government benefits to mitigate those problems, in the short run. The paper answers three research questions. First, how sensitive is program coverage to changes in public expenditure? Second, what are the structural bottlenecks behind poor coverage response? Third, what are the optimal budgetary allocations that could boost the performance of a multidimensional objective function?