Two Ways to Get an AI to Do What You Want
When an AI model doesn’t behave the way you need, there are really only two levers you can pull: change what you ask it (prompting) or change the model itself (fine-tuning). People often reach for the more expensive, more complicated option first, when a better-written prompt would have solved the problem in five minutes. Understanding the actual difference between the two saves both money and frustration.
What Prompting Actually Does
Prompting means giving the model instructions, examples, and context at the moment you ask your question, without touching the model’s underlying weights at all. A well-built prompt can include explicit formatting rules, a handful of example input-output pairs (called few-shot examples), relevant background documents, and step-by-step reasoning instructions. Because nothing about the model changes, prompting is instant, free to experiment with, and fully reversible — you can rewrite a prompt a dozen times in a minute to see what works.
What Fine-Tuning Actually Does
Fine-tuning means continuing to train an existing model on your own dataset of examples, which adjusts the model’s internal weights so the new behavior becomes baked in rather than requested each time. This requires assembling a quality training dataset (often hundreds to thousands of examples), computing resources to run the training job, and a testing process to confirm the fine-tuned model didn’t lose other capabilities in the process. It also locks in a specific snapshot of behavior that won’t automatically improve when better base models are released.
When Prompting Is Clearly Enough
If your problem is “the model doesn’t know the format I want” or “it needs more context about my specific situation,” prompting solves it almost every time. Giving the model two or three examples of the exact output format you want, or pasting in the relevant reference material it needs, fixes the vast majority of real-world frustrations people have with AI tools. This is also far cheaper to iterate on, since you’re not paying for training compute every time you adjust your approach.
When Fine-Tuning Earns Its Cost
Fine-tuning makes sense when you need a very specific, narrow behavior repeated at scale across thousands of calls where the cost of a long, detailed prompt adds up, or when you need the model to reliably use a specialized vocabulary, tone, or format that no amount of prompting fully locks in. It’s also the right call when you have a large, high-quality dataset of exactly the task you want done and need consistency that a prompt — however well-written — simply can’t guarantee across edge cases.
The Practical Takeaway
Start with prompting, always. Spend real effort writing a clear, example-rich prompt before concluding you need fine-tuning — most people who think they need a custom-trained model actually just need a better-written request. Reach for fine-tuning only after you’ve hit a ceiling that better prompts genuinely can’t solve, and you have both the data and the ongoing volume to justify the cost of maintaining a custom model.
