Prompt Engineering

Definition

Prompt Engineering is the practice of systematically designing, testing, and optimizing the inputs provided to an LLM to reliably elicit desired outputs, behaviors, and reasoning patterns. A prompt is everything the model receives as input—system instructions, context, examples, and the user's request. Prompt engineering is the primary lever for controlling LLM behavior in production systems, and in many cases can achieve results equivalent to fine-tuning without the cost and complexity of model training.

Engineering Context

Prompt engineering is an engineering discipline, not an art. Production prompts are versioned, tested against evaluation sets, and changes go through CI/CD pipelines with regression gates. Key techniques include: explicit instruction formatting using XML tags or numbered steps (models follow structured instructions more reliably than prose), few-shot examples (2-5 examples demonstrating the exact format required), chain-of-thought elicitation for complex reasoning tasks, output format specification (describe the exact JSON schema you expect), and negative examples (what the model should NOT do). System prompts are the primary configuration interface for agent behavior—treat them as code: version them, review changes, and test them before deploying to production.

Related Terms