The prompt is everything you feed the model: the question, the instructions, the examples, the tone, the context. Not just the sentence you typed last. The model reads all of it as one block of text and predicts what comes next, which means every part of that block is steering the answer, whether you meant it to or not.
Prompt engineering sounds like a technical specialty, and vendors are happy to let it sound that way. It mostly means wording the ask carefully so the prediction lands where you want it. State the role. Show an example of what good looks like. Say what to leave out. Give the model the context a new hire would need. None of this is code, and all of it moves the output more than most tooling does.
For an operator, the prompt is the cheapest lever in the whole stack. Before commissioning anything custom, the honest first question is how far a better prompt gets you, because the answer is often most of the way. A prompt is also an asset: once one is written and proven for a job, it can be saved, shared, and rerun by anyone on the team, which is how a good prompt quietly becomes a system.
A concrete example. A manager asks a model to write a job posting and gets generic filler. The second attempt includes the company's actual voice, two postings that worked before, the specifics of the role, and what to avoid. Same model, same minute of effort, completely different result. The difference was never the AI. It was the ask.
“Before we build anything custom, let's see how far a better prompt gets us.”
// this page is the full text. the packaged PDF, all fifteen terms, comes with the free library.