Terms Worth Knowing

You've talked to Claude, seen how it works, and watched it use tools. Now it's time to put names on what you've been doing.

Prompt

The message you send to AI.

Explain this report like I'm presenting it to a board.

System Prompt

A hidden instruction the AI receives before your message — written by whoever built the app, controlling tone, role, and rules. You won't write these until later in the course, but every AI you use already has one system prompt running quietly in the background.

Token

A small piece of text — roughly 4 characters in English. That part's intuitive. What's not obvious: tokens aren't just your typed message and Claude's visible reply. The system prompt, your whole conversation history so far, and — if a tool was used — the full raw search results Claude read before answering, all count too. That's why a two-sentence answer can quietly burn through far more tokens than what's on your screen. Most of the cost is processing you never see.

Context Window

The total amount of text the AI can "see" at once, measured in tokens. For Claude, that's around 200,000 tokens. Once it fills up, older messages get dropped — Claude effectively forgets them.

Hallucination

A confident, fluent, but wrong answer. Not a bug, a natural result of how language models work. Reduced by using tools, citations, and better prompts.

Tool

A function Claude can call mid-conversation to get real data or take an action — web search, a spreadsheet, a document. When Claude uses one, it pauses, runs it, reads the result, then keeps answering. You watched this happen last chapter.

Agent

A saved, pre-configured version of Claude with a system prompt and selected tools already set up. Instead of re-explaining context every time, you just launch it. You'll build one later in this course.

LLM — Large Language Model

• Large: billions of parameters;
• Language: works with text;
• Model: a statistical pattern, not a database.

This is what Claude is.