What is the difference between prompt engineering, RAG, and fine-tuning, and when do you use each?

• Prompt engineering — shape the model's behavior through instructions, context, and examples in the prompt. Nothing about the model changes; you just communicate better (system prompts, few-shot examples, output format). Cheapest and fastest; your first resort.
• RAG (retrieval-augmented generation) — at query time, retrieve relevant documents (from a vector store, database, or search) and insert them into the prompt. The model answers from that supplied context. Best when the knowledge is external, private, or changing.
• Fine-tuning — continue training the model on your own examples, adjusting its weights. This bakes in a consistent style, format, or narrow skill. Powerful but costly and static — the knowledge is frozen at training time.

Comparison

Decision guide

• Start with prompting — solve it for free first.
• Need facts the model doesn't know, or that change (docs, prices, internal data)? Use RAG.
• Need a reliable style/format or a specialized task at scale, and prompting isn't consistent enough? Fine-tune.
• These combine: a fine-tuned model with RAG and a good prompt is common in production.

Why it matters

Reaching for the wrong tool is expensive: people often try to fine-tune to add knowledge (which RAG does better and cheaper) or to fix behavior (which prompting handles). Knowing that prompting shapes behavior, RAG supplies knowledge, and fine-tuning changes the model lets you pick the cheapest approach that works — and combine them deliberately rather than by accident.