With computers it has always been clear what you get when you instruct it to do something. Even if it doesn't always feel like that. They are complex systems and sometimes an outcome to a simple instruction can result into unexpected outcome. But technically you always get what you ask for.
With our new A.I. friends the interaction is different. They can't be instructed to do things, not at the level we have used to when interacting with software systems before. The exact same "instruction" might give different results each time. There are no formal rules on how to make them work how you want.
You can only suggest the desired outcome. Adding more details will yield better results, but there is always the element of variance. Even the same suggestions might work differently in different contexts.
This has been a big learning to jump from the formal structures and logic of traditional programming languages into prompting a generative model so that it would produce the desired result. The good thing is it's easy to iterate and tey out different prompt options to reach the outcome you are looking for.