Prompt engineering

Programming an AI through prompts, sometimes called promptware engineering. The result the AI should deliver is set before any prompt is written. Tests then check each version of the prompt against that result. This makes the work the AI does inside a program reliable and efficient. Testing matters most when software sends an AI the same kind of request again and again. There, a weak prompt keeps producing bad results until someone notices. The same prompt often gives a different answer each time, so one good answer proves little. Typing a one-off request into a chat and judging the reply is ordinary prompting.