A framework explaining why LLM performance degrades with task complexity, grounded in information theory, cognitive science, and software engineering; includes metrics and practices for context engineering and human-AI collaboration.