AI code drift causes inconsistent style and architecture across sessions due to LLM non-determinism, diverse training data, and context sensitivity. Learn why this happens and how to mitigate it for better maintainability.