Architecture Decisions
Fine-tuning vs. Prompting
When to teach the model vs tell the model
Prompting gives the model instructions at runtime. Fine-tuning changes the model's weights permanently. Prompting is like giving someone a checklist. Fine-tuning is like training them for months. Most teams fine-tune too early — before they've exhausted what prompting can do.
Why this matters
Fine-tuning costs $10K-$100K+ and takes weeks. Prompting + RAG gets you 80% of the way in days. Fine-tune only when you have proof prompting isn't enough.
Prerequisites