Big Code is all the rage. IDE builders and people who know static analysis see a wealth of opportunity to generate data to study how people program. What better way to identify coders' skills and confusions than to harvest their IDE interactions, compilation attempts, error messages, and code evolution within and across assignments? Won't this also revolutionize programming education? Unfortunately, this tool-builder's dream often leaves instructors without essential information: why are students doing what they do? Unpacking this requires understanding students' design choices in focused contexts, using Little Code and qualitative methods.

This talk will present several studies on how students learn program design through Big and Little Data about Big and Little Code. We'll discuss how students choose program structure, how language choice interacts with program structure, and what misconceptions students have about semantics. We'll also discuss possible implications of machine learning for K-12 and introductory college CS education. The talk aims to raise appreciation and questions about balancing the Big and the Little for improving programming education at many levels.