Prompt Engineering

Prompt engineering is the art of communicating with LLMs effectively. A well-crafted prompt can be the difference between a useless response and a brilliant one — without changing the model or spending a cent on fine-tuning.

The Anatomy of a Good Prompt

The four components of a great prompt: Role (who the model is), Task (what to do), Context (relevant information), Format (how to respond).

Zero-Shot Prompting

Zero-shot means giving no examples — just a direct instruction. Works well for common tasks the model has seen during training.

Classify this review as Positive, Neutral, or Negative:
"The delivery was late but the product itself is great."

Answer:

Few-Shot Prompting

Few-shot means providing examples (demonstrations) in the prompt itself. The model learns the pattern from your examples and applies it to the new input. Dramatically improves accuracy for specific formats.

Classify sentiment (Positive/Negative):

Review: "I love this phone." → Positive
Review: "Terrible battery life." → Negative
Review: "It works but nothing special." → Neutral
Review: "Best purchase I've made this year!" → 

Chain-of-Thought (CoT)

Adding "Think step by step" or showing the reasoning process in few-shot examples dramatically improves performance on math, logic, and multi-step reasoning tasks. The model "thinks out loud" before giving the final answer.

System Prompts

System prompts are instructions given at the start of a conversation that define the model's persona, capabilities, and constraints. They persist across the entire conversation.

from anthropic import Anthropic

client = Anthropic()

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system="""You are a concise Python tutor. Rules:
- Always provide working code examples
- Explain each line with a comment
- Never use advanced libraries without mentioning them
- If unsure, say so""",
    messages=[
        {"role": "user", "content": "How do I read a CSV file?"}
    ]
)
print(response.content[0].text)

Structured Output

Force the model to respond in a consistent format (JSON, XML, markdown table) for reliable downstream processing. This is critical for production systems.

Extract information from the following text and return ONLY valid JSON:

Text: "John Smith, age 34, joined our San Francisco office in March 2022."

Return format:
{
  "name": "string",
  "age": number,
  "location": "string",
  "join_month": "string",
  "join_year": number
}

Common Prompt Anti-Patterns

Frequently Asked Questions