The craft of writing instructions that get an AI to do what you want.
LLMs are extraordinarily sensitive to how you phrase a request. The same task can produce garbage or gold depending on word choice, structure, and what you include. Prompt engineering is the practice of testing, refining, and standardizing those prompts so you get reliable results. It includes things like: giving examples (few-shot), assigning a role ('you are a careful tax accountant'), breaking tasks into steps, and specifying the output format.
Prompt engineering is the cheapest lever you have. Before you fine-tune, before you build agents, before you add RAG - try rewriting the prompt. A better prompt often closes 80% of the quality gap.