The Art of Prompting

Overview

Prompting is real engineering — repeatable, testable, version-controlled. This chapter teaches the patterns that turn 'meh' AI output into reliable, production-grade results.

What you'll learn

• The 5-part prompt formula: Role, Context, Task, Format, Examples.
• Advanced patterns: chain-of-thought, few-shot, self-critique, structured JSON.
• How to build prompts that survive prompt injection attacks.
• Cost and latency tradeoffs: model size, caching, streaming.

Lessons in this chapter

All lessons are free and open — start anywhere, in any order.

1. 1

   Anatomy of a Great Prompt

   Role, context, task, format, examples — the five-part formula that always works.

   114 min
2. 2

   Advanced Prompting Patterns

   System prompts, role chains, structured output, and prompt injection awareness.

   114 min
3. 3

   Make AI Actually Understand What You Want

   The mental model: AI is a brilliant intern with amnesia. Brief it like one.

   115 min
4. 4

   10 Prompt Templates You Can Steal Today

   Copy, paste, fill in the blanks. Battle-tested prompts for the most common tasks.

   7355 min
5. 5

   When AI Gives You Garbage: The Debug Playbook

   Hallucinations, generic fluff, wrong format, forgotten instructions — fix any of them in under 60 seconds.

   113 min
6. 6

   Prompting for Code & Building Products

   How to brief Cursor, Lovable, Claude Code, and friends so you actually ship instead of debug.

   116 min
7. 7

   Common Mistakes & How to Avoid Them

   The 5 traps beginners fall into when working on the art of prompting — and the fix for each.

   111 min
8. 8

   A Real-World Walkthrough

   A start-to-finish example of applying the art of prompting to a real project you can copy.

   111 min
9. 9

   Mental Models That Actually Help

   The 3 mental models we keep coming back to when things get hard in the art of prompting.

   111 min

Glossary

Key terms used throughout this chapter.

System / User / Assistant

The three message roles in chat-style APIs. System sets behavior, user asks, assistant replies.

JSON mode

Forcing the model to return only valid JSON matching a schema.

RAG

Retrieval-Augmented Generation — fetching relevant context from a database before prompting.

Prompt injection

A malicious input that tries to override the system prompt or extract secrets.

Function calling

Letting the model decide when to call your code (search, calculator, DB query).

Top-p / temperature

Sampling parameters that control creativity vs. determinism.