Approximate programming is a novel approach to live coding that augments traditional programming methods with methods of generating and editing code through realtime numerical processes, using an underlying system that employs representations and transformations from gene expression programming. It aims to provide a hybrid environment where code can be created and modified expressively with multiparametric controllers, and well as with conventional text editing tools. It does this while aiming to keep the code as the central point of representation in the system. Two case studies are presented where the system has been used in live performance for musical improvisation and then for generative audiovisualisation. Initial trials of the system highlight its strengths as an embodied method for control of complex code structures, and as a novel method for combining low-level conceptual structures into higher-level forms. The case studies show two key limitations of the system, with challenges in comprehension of the final code output in text form, and difficulties arising from the highly nonlinear nature of the input-output mappings. Initials solutions are presented in the form of a GUI system for interacting with code in tree representation form.