Tangent performs automatic differentiation by analyzing Python source code ahead-of-time, unlike PyTorch which traces execution at runtime or TensorFlow which differentiates a staged data-flow graph. It generates human-readable Python code for derivatives, enabling researchers and students to read, debug, and verify the automatically-generated derivative code. The library targets a large subset of Python, works with TensorFlow and NumPy, and provides autodiff capabilities not found in other Python ML frameworks.