Prompt engineering is not a collection of tricks, it is iteration: you write it, try it on a few real cases, see where it drifts from what you wanted, and write the difference in. The gap between hobby use and professional use is not in the vocabulary, it is in whether you run the same prompt several times, on several cases, before you let it go live.
Why does it matter? Because in a live process a prompt is code: when you are labelling a hundred orders, what interests you is not that the answer looked good once, but how often it gets one wrong. A prompt therefore deserves the same treatment as a piece of code — versioned, with examples, and measured.
The most common misconception is that the discipline is over because models got smarter. Partly true: a lot of the old tricks are redundant. What has not gone away is the design of the context — what goes into the model's window next to the question — and by now that is the bigger half of the work.
The five building blocks and the practical iteration are the subject of The Anatomy of a Working Prompt, a post of its own. Directly related terms: context engineering, meta-prompting and prompt library.