What's the Deal with Context? (And Why AI Has ADHD)

What's the Deal with Context? (And Why AI Has ADHD)

You know that friend who remembers your birthday, your favorite snack, AND that embarrassing thing you did in seventh grade? Yeah, AI isn't that friend.

AI is more like the friend who remembers your birthday but asks "wait, who are you again?" every time you start a new conversation. And the worst part? It just did something brilliant thirty seconds ago.

Let me set a scene. You're using ChatGPT to help you with history homework. You ask it to explain World War I. It does a beautiful job—vivid, clear, packed with dates and details. You follow up: "What about the role of colonial troops?" Nailed it. "How did technology change warfare?" Brilliant. "Wait, going back to your first answer, you said something about the Balkans..."

And suddenly, the AI goes blank. "I apologize, but I don't see any previous context about the Balkans in our conversation."

You're sitting there staring at the screen like, *bro. You literally just talked about it.*

Welcome to the central mystery of modern AI: it can read a 300-page novel in under a second and summarize it flawlessly, but it can't remember what YOU said three messages ago. It's like having a photographic-memory genius with no short-term memory at all.

And the reason for this isn't laziness, stupidity, or bad engineering. It's actually one of the deepest architectural choices in how AI works—and understanding it is the difference between someone who uses AI like a confused tourist and someone who uses AI like a confident wizard.

In this chapter, we're going to crack open the single most important concept in working with AI: context.

By the end, you'll understand:

• What "context" actually means in AI-land (and why it's different from what humans mean by context)
• Why every AI has a "context window"—an invisible, limited whiteboard it carries around
• Why AI behaves like it has ADHD: brilliant in short bursts, hopelessly lost in long ones
• The four superpowers people use to break through this limitation: long context, compression, summarization, and retrieval
• Why understanding this turns you from an AI user into an AI collaborator

Let's go.

What Even Is "Context"? (And Why It's Not What You Think)

Let's start with the most basic question. When humans say "context," we mean a million things at once.

"Read it in context." "Don't take that out of context." "What's the context here?" "He has no context for that reference."

Human context is fuzzy. It's the weather, your mood, what you ate for breakfast, that song stuck in your head, the political climate, your relationship with the person talking, the last twelve books you read, and a thousand invisible threads that all braid together to make meaning.

AI context is way more boring—but also way more powerful when you understand it.

AI context is just the text the AI can currently see.

That's it. That's the whole definition.

When you type a message to ChatGPT, Claude, Gemini, or any other chatbot, the AI doesn't magically know what you're thinking, what you did yesterday, or what your teacher said last week. The only information it has is: 1. The text you've typed in this conversation 2. A system prompt that gives it instructions like "You are a helpful tutor" 3. Maybe a few examples or background documents loaded behind the scenes

Everything else? Gone. Invisible. Not part of the picture.

This is the first big mental shift. When you're talking to an AI, you're not talking to a person who knows your whole life story. You're talking to a super-smart entity that can only see what's written on the page in front of it at that exact moment.

Think of it this way: if you've ever worked as a tutor, you know the difference between tutoring a kid you've known for years and tutoring a kid who just walked in cold. With the long-term kid, you can say "remember when we did fractions?" and they'll be like "oh yeah." With the new kid, you have to explain everything from scratch.

AI is always the new kid. Every single conversation.

Quick aside: this is starting to change with "memory" features that some AI tools are adding, where the AI can store facts about you across conversations. But those are bolt-ons we'll talk about later. The core architecture is still the same.

The Conversation Is the Context

So if context is "what the AI can see right now," then every word you and the AI have said in the current conversation is part of that context.

Let's trace what happens in a typical chat:

Turn 1 (You): "Hey, can you help me understand photosynthesis?" Turn 1 (AI): "Absolutely! Photosynthesis is how plants convert sunlight into energy..." Turn 2 (You): "Cool, what about how it relates to cellular respiration?" Turn 2 (AI): "Great question! Photosynthesis and cellular respiration are basically opposite processes..."

After Turn 2, the AI's context includes:

• Your Turn 1 message
• Its Turn 1 response
• Your Turn 2 message
• The system prompt (invisible to you)
• Whatever internal instructions it was given

All of that combined is what the AI is "looking at" when it generates Turn 2's response. It's holding all that text in its head, in a sense, and using it to inform what it says next.

This is why the conversation feels coherent. The AI isn't just answering your latest question—it's answering based on everything you've said and everything it's said, all at once.

But—and here's where it gets tricky—there's a limit.

The Invisibility Problem

One of the weirdest things about AI conversations is that the context is invisible. You don't see the AI "looking at" the previous messages. You don't see it pulling up notes. You don't see the buffer filling up.

From your perspective, you're just typing messages and getting responses. The whole machinery is hidden.

But behind the scenes, every word of the conversation is sitting in a buffer. Every. Single. Word.

Let's say you ask the AI to write a 1,000-word essay. Now the context includes your prompt (maybe 30 words) plus its 1,000-word response. That's around 1,030 words the AI is "looking at" to answer your next message.

Ask it to write another 1,000-word essay. Now we're at 2,030 words. And another. Now we're at 3,030.

You see where this is going.

There's a finite amount of text the AI can hold in mind at once. That limit is called the context window, and understanding it is the key to understanding everything else.

The Invisible Whiteboard: Meet the Context Window

Imagine the world's smartest person is sitting across from you at a café. They're brilliant. They can solve differential equations in their head, quote Shakespeare, and explain quantum mechanics with perfect clarity.

But here's the catch: they have a small whiteboard.

Not a giant classroom chalkboard. Not a notebook. A small whiteboard. The size of a piece of paper, maybe. It can only hold so much writing before you have to erase something to make room for new writing.

That's the context window.

When you say something to this brilliant person, they write it on the whiteboard. When they answer, they write that too. When you follow up, it goes on the whiteboard. The whiteboard is filling up.

At some point, the whiteboard runs out of space. To make room for your new question, they have to erase the oldest stuff. And once it's erased, it's gone. They can no longer "see" it. They can't reference it. They don't even know it was ever there.

That's exactly what happens inside an AI.

The context window is the AI's whiteboard. It holds:

• Your messages
• The AI's responses
• Any system instructions
• Any background documents you uploaded

And it has a hard size limit. Different AI models have different whiteboard sizes. Some are tiny (like the original GPT-3, which could only hold about 4,000 tokens—that's roughly 3,000 words). Some are huge (modern models like Claude can hold hundreds of thousands of words, basically entire books).

But all of them have a li