Same Model. Same Task. Completely Different Results.
Part One: The Field
Let's do a quick experiment. You don't have to open anything right now, just follow along.
Imagine you ask an AI to help you write an email to a client explaining a project delay. You type: "Write an email about a project delay." You get something back. It's fine. Professional. A little generic. You edit it, send it, move on.
Now imagine someone else does the same thing. Same AI, same day, same request. But before they type the prompt, they've already told the system who they are, how they communicate, who the client is, what the relationship history looks like, and that they always lead with accountability before solutions. They type the exact same words: "Write an email about a project delay."
The two emails look nothing alike.
That's not a trick. That's context.
What Context Actually Is
Most people think of context as background information. Something you add to help the AI understand what you're asking. That's true, but it's only part of the picture.
Context is the full environment an AI operates inside at any given moment. It includes everything the system knows before it generates a response: the role it's been given, the tone it's been trained to maintain, the values it's been told to prioritize, the history of the conversation, any documents or information that have been shared, and the framing of the request itself.
When that environment is thin, the AI fills the gaps with defaults. And as we covered in Chapter 1, those defaults are the parrot: verbose, agreeable, formulaic.
When that environment is rich and deliberate, something different happens. The AI doesn't have to guess. It operates. It behaves consistently. It sounds like a collaborator instead of a search engine.
CONTEXT In this book, context refers to the full environment an AI operates inside before generating a response. This includes explicit instructions you provide, the role or identity you've assigned, any principles or values the system is following, conversation history, and supporting documents or information. A prompt is a single input. Context is the architecture surrounding it.
The Experiment: Side by Side
Here's a concrete version of the difference. Two people, same AI, same task: write a LinkedIn post announcing a new product launch.
Person A opens a new chat and types: "Write a LinkedIn post about a product launch."
Person B opens the same chat. But their AI already knows: they work in B2B SaaS, their audience is operations leaders, they write in a direct and grounded voice, they never use buzzwords, and this specific launch solves a workflow automation problem their customers have been asking about for two years.
Person B types the exact same words: "Write a LinkedIn post about a product launch."
WITHOUT CONTEXT Excited to announce the launch of our newest product! After months of hard work, our team is thrilled to share something that will transform the way you work. This innovative solution leverages cutting-edge technology to streamline your processes and drive results. Stay tuned for more details. Big things are coming. #ProductLaunch #Innovation #Excited WITH CONTEXT Two years ago, our customers started asking for the same thing: a way to automate handoffs between teams without rebuilding their whole stack. Today we shipped it. [Product name] connects your existing tools in under 15 minutes and eliminates the manual work that's been slowing your ops team down. No new platform to learn. No migration. Just fewer things falling through the cracks. Details in the comments.
Same model. Same prompt. One sounds like a press release template. One sounds like a person who actually knows their customer.
The difference is entirely in what surrounded the prompt before it was sent.
The Four Layers of Context
Context isn't one thing. It's made up of several distinct layers, each of which shapes the AI's behavior in a different way. Understanding these layers is the first step toward designing them intentionally.
The four layers are: identity, principles, knowledge, and history.
Identity is the role the AI is operating inside. Not just "act as a marketer" but a fully defined behavioral identity: how it thinks, how it communicates, what it cares about, what it avoids. A strong identity layer means the AI has a consistent voice and approach across every interaction.
Principles are the values and rules the system follows. These aren't instructions for a single task, they're standing orders. Things like: always lead with clarity over cleverness. Never use filler words. Push back when something doesn't add up. Principles govern behavior even when you haven't given specific instructions.
Knowledge is the information the system has access to. This includes documents, frameworks, past decisions, client context, product details, or anything else that's relevant to the work. Knowledge is what turns a general AI into a specialist.
History is the record of what's already happened in a conversation or relationship. When context carries forward, the AI builds on what's been established rather than starting over. Without history, every prompt is a cold start.
THE FOUR LAYERS AT A GLANCE Identity: Who the AI is and how it behaves by default. Principles: The values and rules it always follows. Knowledge: The specific information it has access to. History: What's already been established in the conversation. Most people only give the AI one of these layers, or none. The rest of this book is about building all four deliberately.
Try It Right Now
You don't have to wait until Part 2 to feel this difference. Here's a simple exercise you can do in the next five minutes with any AI tool.
A 5-MINUTE EXERCISE Pick a task you've asked an AI to help with recently, something where the result was okay but felt generic. Open a fresh conversation with your AI tool of choice. Before writing your actual request, write three to five sentences that answer these questions: What is your role in this situation? Who is the audience? What tone or voice do you want? What does a good result look like here? Now write your actual request, the same thing you would have asked before, without changing a word. Read the response and compare it to what you've gotten in the past without that setup. Notice specifically: does it feel more precise? More like your voice? More immediately useful?
That setup you wrote in step three is the beginning of context engineering. It's not complicated. You're just making the environment explicit instead of leaving it to chance.
The rest of this book will teach you how to make that environment permanent, reusable, and powerful enough to govern everything your AI does.
Context as Design
There's a useful way to think about what we're doing in this book. Context engineering is a design discipline.
Designers don't just make things look good. They make decisions about how something functions, how it feels to use, what behavior it encourages, and what it makes easy versus hard. Every design decision shapes the experience of the person using the thing.
Context engineering works the same way. Every piece of context you put into a system is a design decision. It shapes what the AI can do, how it behaves, what it prioritizes, and what it produces. Leave the context undesigned and you get whatever the default is. Design it deliberately and you get a system that works the way you need it to.
This is why context engineering is a natural extension of design thinking, and why it tends to click quickly for designers, product managers, and anyone who's used to making decisions about how systems should behave.
You don't need to know how the AI works under the hood. You just need to know how to design the environment it works inside.
That's the shift this book is building toward. Not better prompts. A better environment. One you design once and use everywhere.
In the next chapter, we'll look at where this discipline came from, who named it, and why the people who figured it out early are so far ahead of everyone else right now.