We study how artificial intelligence affects welfare in general equilibrium when production requires both finding and solving problems. Unsolved problems are unexploited opportunities or unmet needs that agents find in a congested sector. Problems are solved in layered hierarchies in which agents can ask for help. Firms may adopt a replicable AI agent in either activity. AI deployed in solving problems splits the human knowledge hierarchy: greater AI capability affects the layers above AI, whereas lower cost affects the layers below it. Either form of progress can lower welfare by reallocating solvers to the congested finding sector. A tax on finding corrects this distortion; when this is not possible, the optimal tax on AI can be positive or negative, depending on its effect on the congestion externality.