Prompt engineering is often thought of as simply re-phrasing your Google search sentence a few times. Deep prompt engineering for generative AI is entirely different. It’s designing a series of instructions, constraints, and examples to guide the model towards a desired–often multi-faceted–result. A simple example: “Gemini, make your answer sound like Mark Twain wrote it, complete with his sly humor and distinctive vernacular.”

Deep prompt engineering is a blend of linguistic skill, creativity, and an understanding of the model’s architecture. The team and I dove in, writing complex scenarios–often using CLI (Command Line Interface) to enter them. We also employed multi-shot prompting; we’d give Gemini a few examples of input-output pairs to show the exact pattern we wanted it to follow.