Watch: Prompt Engineering
Writing and structuring model input so the output is reliable enough to use. Where the problem is what the model is given rather than how it is asked, the discipline is context engineering.
Transcript
Prompt engineering is the art of structuring your input to get the most reliable output from an AI model. It is about more than just asking a question. It involves the instructions you write, the examples you provide, and even the order of your text.
Several key techniques can make a big difference. Few-shot prompting gives the model worked examples to follow without needing extra training. Chain-of-thought prompting asks the model to show its step-by-step reasoning, which greatly improves accuracy on complex tasks. You can also control the output format with a strict schema to prevent parsing errors, or lower the temperature setting to reduce variation when there is only one correct answer.
Where you place your instructions matters, too. AI models pay closer attention to the very beginning and the very end of a long text. Because of this, it is usually best to place your instructions after any long reference material.
But prompt engineering has its limits. It cannot supply knowledge the model does not already have. For that, you need database grounding, not just better phrasing. Prompts are also highly model-specific, meaning a prompt tuned for one system might degrade on the next version. Finally, any prompt that includes untrusted text is vulnerable to prompt injection, where the system is manipulated into ignoring its original instructions.
Ultimately, if the problem is about what information the model is given rather than how it is asked, you are dealing with context engineering.