Human culture is remarkable for its ability to accumulate knowledge and innovations across generations. Yet theories of cultural evolution often explain this capacity through relatively simple processes: individuals innovate, others copy successful behavior, and useful variants accumulate over time. This leaves much of the cognitive machinery that makes cumulative culture possible unspecified. In this talk, I will present three projects that put this machinery back into theories of cultural evolution. First, I will show that people learn how to learn from others, flexibly acquiring social learning strategies through experience. Second, I will show that innovation is not simply random variation, but is guided by semantic knowledge that structures the search for new solutions. Finally, I will show that language enables cumulative improvement by transmitting underlying knowledge that cannot be readily recovered from observing behavior alone. Together, these findings suggest that the cumulative power of human culture depends on cognitive processes that shape whom we learn from, how we generate new solutions, and what knowledge we transmit to others.