Prompting¶

The prompt is your primary interface to an LLM. Prompting well is the highest-leverage, lowest-cost skill in AI engineering — and the first one worth mastering.

Overview¶

Before you reach for fine-tuning, RAG, or agents, you reach for the prompt. Small changes in how you ask can be the difference between an unusable feature and a shippable one — at zero additional infrastructure cost. This section takes you from writing clear instructions to getting reliable, structured, tool-using behavior.

What you'll learn¶

Prompt Engineering

The core techniques: clarity, examples, chain-of-thought, and iteration. Start here.
System Prompts

Set persona, rules, and format once — for the whole conversation.
Structured Outputs

Get reliable JSON you can parse, validate, and build on.
Function & Tool Calling

Let the model do things: search, calculate, call your APIs.

Learning Objectives¶

By the end of this section you will be able to:

Write clear, well-structured prompts that reliably get good results.
Use system prompts to control behavior across a conversation.
Force structured, machine-readable output and validate it.
Give a model tools and handle its tool calls safely.

The prompting mindset¶

flowchart LR
    W[Write a clear prompt] --> T[Test on real inputs]
    T --> M{Good enough?}
    M -->|no| I[Inspect failures<br/>refine]
    I --> W
    M -->|yes| E[Lock it in<br/>+ add an eval]

Prompting is empirical. Write, test on realistic inputs, inspect failures, refine. When it's good, protect it with an evaluation so future changes don't regress it.

Escalation ladder

Reach for the cheapest tool that solves the problem, in this order: prompt → structured output → tools → RAG → fine-tuning. Most problems are solved before you get to the end.