How do people adopt new tasks in evolving fields? Despite the proliferation of data on occupations and tasks such as O*NET, the limited resolution of occupation-derived task data and information about individuals’ in-job activities limits our ability to understand the learning process. In this paper, we derive a fine-grained and hierarchical taxonomy of software development tasks using data from tens of millions of posts on Stack Overflow, the largest online Question and Answer community for software. In particular, we demonstrate that elemental programming concepts are meaningfully interconnected by co-occurrences within posts and user posting histories. Communities within these networks define tasks. We show that individuals create higher quality content as they gain experience in a task, and are much more likely to learn tasks related to ones they already know. Our findings suggest that fine-grained digital trace data can be used to both define meaningful task landscapes and to understand the dynamics of individual learning.