I've found that something very similar to those "inefficient" processes works incredibly well when applied to LLMs. All of those processes are designed to allow for seamless handoff to different people who may not be familiar with the project or code which is exactly what an LLM behaves like when you clear its context.