Great breakdown this really captures the shift toward simplicity, caching, and system-level thinking in LLM design rather than overcomplicating prompts and architecture.