Prompt engineering is the practice of writing, testing and adjusting the instructions an AI model receives until it answers well consistently. It does not change the model: it changes what you ask. The core is giving context, an example and the expected format, then checking the result against real cases.
A question from the app. Pick an answer.
You ask for a summary and get something long and generic. What do you change first?
A role can change the tone, but the AI still does not know who it is for or how long you want it.
That is what was missing: the context and the shape of the result, like "three bullets for my boss".
Each try can vary, but with the same vague request the problem usually comes back.
A prompt is the text you give the AI. Prompt engineering is the work of improving it on purpose: write it, see what comes back, change one thing and test again. Context engineering looks at the bigger picture: everything the model sees at each step, not just your request.
Vague: "Summarize this email." It does not say who it is for or how long it should be. Clearer: "Summarize this email for my boss, who has one minute. Context: it is from a supplier announcing a delay. Format: three bullets, the first one with the new date. Example bullet: 'Delivery moves to Nov 15'." What changed is not a magic phrase: it is three pieces, context, format and an example.
In a chat, if the answer is not useful, you fix it on the spot. A prompt inside an app runs thousands of times on texts you never saw. Working once is no longer enough: you collect 20 real cases (odd ones included), write down what should come out, and test every prompt change against all of them. If you fix one and break three, you will see it.
A good prompt reduces mistakes but does not remove them: the model can still make things up. And if the prompt handles other people's text, like emails or web pages, someone can hide instructions in it: that is prompt injection.
Writing and testing prompts is useful in software development, product work and support. What matters is explaining what you need and checking the results against real cases.