Really well-written piece on the evolution of prompt engineering! We are moving from magic incantations that make no sense to a regular human, to intricate no-gradient stacks around an LLM that avoids under-specifying a task.