What is prompt engineering?

Prompt engineering is the practice of designing inputs to AI language models to get accurate, useful, and consistent outputs. Unlike traditional programming where you write deterministic code, prompt engineering involves crafting natural language instructions that guide a probabilistic model. Good prompt engineering reduces hallucinations, improves output quality, and makes AI tools reliable enough for production workflows.

// Structured prompt template for code review
const prompt = `
You are a senior software engineer performing a code review.

## Context
Language: ${language}
Framework: ${framework}

## Code to Review
\`\`\`${language}
${code}
\`\`\`

## Review Criteria
1. Security vulnerabilities (injection, XSS, SSRF)
2. Performance issues (N+1 queries, memory leaks)
3. Error handling completeness
4. Type safety and null checks

## Output Format
For each issue found, provide:
- **Severity**: Critical / High / Medium / Low
- **Line**: affected line number
- **Issue**: description
- **Fix**: suggested code change
`;

Core techniques

The most effective prompt engineering techniques include: Role Assignment (giving the model an expert persona), Few-Shot Examples (providing 2-3 input/output pairs), Chain of Thought (asking the model to reason step by step), Structured Output (specifying exact response format like JSON or Markdown), and Constraint Setting (defining boundaries like max length, included/excluded topics, and style). Combining these techniques yields the best results.

System prompts vs user prompts

System prompts define the AI's persona, capabilities, and rules — they persist across the conversation. User prompts contain the specific request or data for each turn. OpenAI and Anthropic support dedicated system message fields. For Gemini, system instructions are prepended as the first user message. Separating system and user content makes prompts more maintainable and reusable across different requests.