Few-Shot Examples

Few-shot learning provides the model with concrete examples of desired input-output pairs directly in the prompt, guiding its behavior through demonstration rather than instruction alone. This technique is one of the most reliable ways to improve output quality for formatting, tone, and domain-specific conventions that are difficult to describe in words but obvious when shown, and it is particularly useful in agentic coding for demonstrating exactly what a correct tool call should look like in your system. The key tradeoffs are coverage versus cost: examples must represent edge cases well enough to steer behavior, but each example consumes context window tokens that might otherwise hold task-relevant information, so selecting the most representative few examples matters more than using many.