To make a general model fit a specific task, you can either adjust its internal weights by training it further, called fine-tuning, or simply give it careful instructions and examples in the prompt. Prompting is quick and cheap to iterate; fine-tuning bakes behaviour in more deeply.
The choice depends on scale and stability. For a handful of varied tasks, good prompts usually suffice. For a narrow, high-volume task where consistency matters, fine-tuning can be worth the extra effort and data it requires.