Station 6: An Intro to Context Engineering

What Is Context Engineering?

Context engineering is the practice of deliberately providing AI with the information it needs to give you useful responses. It's less about how you phrase your question and more about what information you include.

"Context engineering is focused less on clever phrasing and more on bringing the right information (in the right format) to the LLM."

That quote was my breakthrough moment. I'd been trying to write "perfect prompts" — tweaking my wording, using special phrases I'd read about online. But the real skill is simpler and harder: giving AI the information it actually needs.

Grace put together a comprehensive guide on context engineering that I keep bookmarked. I'm still learning from it, but I want to share the key concepts that helped me the most.

📚 Full Reference: Grace's Context Engineering Guide — everything I know about this topic came from here.

The Three Types of Context

Grace's guide explains that AI coding assistants get context from three sources. Understanding these helped me know what I need to provide versus what AI already has.

1. Calculated Context (Automatic)

This is context AI gets automatically, without you doing anything:

• Files open in your editor tabs — AI can see what you're working on
• Your current file and cursor position — AI knows where you are in the code
• Recently edited files — AI has some memory of your recent work
• Your project's package.json — AI knows your dependencies (if present)

What this means for you: If you want AI to consider a specific file, open it in a tab. Close files you're not working on to reduce noise.

2. Implicit Context (Inferred)

This is context AI figures out from patterns in your code:

• File extensions — .js, .ts, .py tell AI what language you're using
• Coding patterns — Do you use arrow functions? Semicolons? AI notices
• Import statements — AI can see what libraries you're using
• Naming conventions — camelCase, snake_case — AI tries to match your style

What this means for you: Be consistent in your code style, and AI will follow along. Inconsistent patterns confuse both AI and humans.

3. Explicit Context (You Provide)

This is where context engineering happens — the information you deliberately give AI:

• Comments describing intent — What is this code supposed to do?
• JSDoc documentation — Function signatures, parameter types, return values
• Clear function names — validateUserEmail() tells AI more than check()
• Your prompt itself — The question you ask and the context you include

What this means for you: This is what you control most directly. The rest of this station focuses on providing better explicit context.

The Five Questions Framework

Whether I'm debugging, learning something new, or building a feature, I've learned to answer these five questions in my prompts. They work for almost any AI interaction.

These Questions Work for Everything

I use this framework for:

• Debugging: What should happen vs. what's happening, plus constraints
• Learning: What I'm trying to understand, what I already know, what's confusing me
• Building features: What I want to create, what exists already, what requirements I have
• Refactoring: What I want improved, what must stay the same, why I'm changing it

Not every prompt needs all five — but the more complex the task, the more questions I answer.

What AI Can and Can't Do

Understanding AI's strengths and limitations helps me know when context engineering will help most — and when I need a different approach entirely.

HAP's Takeaway: AI Teaches Best When You Ask the Right Questions

Here's what surprised me most: AI is an incredible teacher — but only when I ask the right way. When I asked "why does condition order matter?" AI gave me a step-by-step explanation I could actually understand. That's teaching at its best: immediate, focused on exactly what I needed.

But when I asked vague questions like "make my code better," AI would make changes I didn't understand — sometimes breaking things it didn't know were important. The difference wasn't AI's capability; it was my prompt quality.

Prof. Teeters put it this way: "AI can teach you anything you ask about — but it can only answer the question you actually asked. The skill isn't using AI; it's asking the right questions." 🟠

Writing Better Prompts

Let me show you how adding context transforms AI responses. These examples use the Five Questions Framework.

Example 1: Learning a Concept

Explain closures

Explain closures for a JavaScript beginner in week 3 of learning.

What I know:
- Functions, parameters, return values
- Variables and scope (global vs local)
- I just learned about function factories

What's confusing me:
- Why does the inner function "remember" the outer variable?
- When would I actually use this?

Please use simple examples with robot-themed variables.

Example 2: Understanding Code

What does this code do?

if (value) {
    return true;
}

What does this code do, specifically with different input types?

if (value) {
    return true;
}

I'm confused about what happens when value is:
- 0 (a number)
- "" (empty string)
- null or undefined
- false (boolean)

This is validation code for a form. I need to understand
which inputs get rejected and which pass through.

Example 3: Getting Help with a Feature

Add form validation

Help me add validation to a nickname input field.

Requirements:
- Reject empty strings, null, undefined
- Accept any non-empty string (including spaces)
- Accept 0 and false (this field is only for nicknames,
  but other fields accept numbers)
- Return true if valid, false if invalid

Current code:
function validateNickname(value) {
    // Need help here
}

Keep it simple - vanilla JavaScript, no libraries.

Project-Level Context (A Preview)

Grace's guide mentions something I haven't set up yet, but want to learn: project-level instruction files. These give AI persistent context about your entire project.

What Is a Project Instruction File?

A file like .github/copilot-instructions.md that AI reads automatically for every request. It contains:

• Project overview: What is this app? Who's it for?
• Tech stack: What frameworks, libraries, and patterns do you use?
• Coding conventions: Semicolons? Arrow functions? Naming patterns?
• Data structures: What do your objects look like?
• Common patterns: How do you handle errors? Async code?

Why This Matters Later

Right now, I'm adding context to each prompt manually. That's fine for learning. But eventually:

• I'll be working on bigger projects where explaining context every time is tedious
• I'll want AI to follow my team's conventions automatically
• I'll have complex requirements that shouldn't need repeating

That's when project-level context becomes valuable. For now, I'm focusing on mastering the Five Questions Framework.

When NOT to Use AI

Context engineering makes AI more helpful, but there are still times when AI isn't the right tool. Understanding when NOT to use AI is part of using it well.