Fine-tuning adjusts a model's internal weights using your own training examples, teaching it a style, format, or specialized skill. RAG, by contrast, leaves the model unchanged and instead feeds it relevant context at query time.
RAG is generally the better fit when your goal is grounding answers in current, changing information — documentation, policies, product catalogs — since updating a vector database is far cheaper than retraining a model.
Fine-tuning tends to make more sense when you need the model to consistently follow a particular tone, structure, or domain-specific reasoning pattern that's hard to express through prompting alone.
In practice, many production systems use both: a fine-tuned or well-prompted model for behavior, combined with RAG for factual grounding.