Prompt Engineering Basics

Introduction

Prompt engineering is the discipline of writing instructions that reliably steer an LLM. In agents, every step (planner, tool selector, summariser, critic) is a prompt — get them right and your agent works; get them wrong and it spirals.

Beginner analogy: prompts are job briefs for a new contractor — vague briefs get vague work; tight briefs with examples and acceptance criteria get great work.

Understanding the topic

Core concepts:

• Role: who is the model? ('You are a senior support agent.')
• Task: what to do, concretely.
• Format: JSON? Bullets? Schema?
• Constraints: length, tone, do's/don'ts.
• Examples: 1-3 shots dramatically improve quality.

Syntax reference

Visual workflow / architecture:

┌─────────────────────────┐
│ System: role + rules    │
├─────────────────────────┤
│ Few-shot examples       │
├─────────────────────────┤
│ User task               │
├─────────────────────────┤
│ Output schema           │
└─────────────────────────┘
            │
            ▼
       LLM call

Real-world use

Every Copilot suggestion, every ChatGPT custom GPT, every Claude project is a prompt-engineered system.

Best practices

• Iterate prompts like code — track versions and diffs.
• Run prompts on a fixed eval set before deploying changes.

Common mistakes

• Tweaking prompts in production without an eval harness — silent regressions.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Five components of a good prompt?
• Q2. Why version your prompts?
• Q3. Scenario: outputs drifted after a prompt change. What process do you follow?