Glossary · From the book

Glossary

From Cleared For Takeoff: The Complete Edition

The full glossary for Cleared For Takeoff: The Complete Edition. The printed book carries the terms its chapters use.

Every industry has jargon. AI has a lot of it. Here is the plain translation.

This glossary defines AI terms the way they actually matter to you as a creator and business owner. Not the technical definition, but the operational one. The definition that helps you use the thing rather than just understand it.

Mostly in aviation language, because this is Heartliner One and we have a theme. Where the metaphor stopped working I stopped forcing it. Which is also a lesson.

Adaptive FluencyStaying Airborne.

The core skill this book is about: picking up new tools quickly, dropping what no longer serves, and staying airborne without having to land every time the instruments are updated. (See: the Captain’s Log.)

Agentic AIAutopilot Mode.

AI that executes multi-step tasks independently, making decisions along the way without waiting for you to approve each one. Your Night Shift crew operates agentically. You set the flight plan; they fly it.

AI HallucinationInstrument Failure.

When AI confidently presents false information as fact. The instruments look normal, the readings look right, but the altitude is wrong. Always cross-check. (See: Chapter 14, You Are the Judgment.)

AI ModelAircraft Engine.

The underlying technology that powers an AI tool. The engine is the model underneath; the chat interface you type into is the cockpit built around it. ChatGPT swaps engines over time, and so does everyone else. Learn the cockpit, not the engine serial number.

API (Application Programming Interface)Air Traffic Control Frequency.

The communication channel that lets different software systems talk to each other. When Forge connects your website to your payment processor, it’s tuning into the right API frequency.

AutomationAutopilot.

A system that performs tasks without manual input once configured. Not the same as AI: your dishwasher is automated but not intelligent. Make scenarios are automated. Ember is automated AND intelligent.

ChatbotIntercom System.

An AI interface designed for conversation. The simplest version of AI interaction: you talk, it talks back. Most people’s first encounter with AI is through a chatbot.

Claude / Claude Code

See Archer and Forge in the Crew Manifest. Claude is the engine. Archer and Forge are what happens when you give that engine a name, a role, and a job description.

Context WindowCargo Hold Capacity.

The amount of information an AI can hold in its working memory during a single conversation. Bigger context window = more documents, more history, more nuance. Current flagship windows reach a million tokens, several books’ worth. When it runs out, the AI starts forgetting what you said at the beginning.

Fine-TuningEngine Modification.

Training an existing AI model on your specific data to make it better at your specific tasks. Expensive, technical, and usually unnecessary for most creators. Custom instructions and good prompting get you 90% of the way there.

LLM, GPT and TransformerThe Engine.

An LLM (Large Language Model) is the class of engine behind ChatGPT, Claude, and Gemini, trained on enormous amounts of text to predict and generate language. GPT (Generative Pre-trained Transformer) is one make of that engine, the one behind ChatGPT. The Transformer is the design underneath most of them, published in a 2017 paper with the beautifully confident title ‘Attention Is All You Need.’ It was right. You don’t need to understand the engineering any more than you need to understand turbine blade metallurgy to fly a plane.

Machine LearningFlight School for AI.

The process by which AI systems learn from data rather than being explicitly programmed. The AI looks at millions of examples and figures out the patterns. Nobody tells it the rules; it derives them.

MCP (Model Context Protocol)Universal Docking Port.

A standard that lets AI tools connect to external services and data sources. When Archer connects to your Google Drive or your calendar, he’s using MCP.

Multi-ModalMulti-Engine Rated.

An AI that can process more than one type of input: text, images, audio, and video. Most current flagship models are multi-modal. They can read your document, look at your image, and listen to your voice note. Not all AI can do this.

Natural Language Processing (NLP)Translation System.

The branch of AI that deals with understanding and generating human language. It’s the reason you can talk to AI in plain English instead of code.

Neural NetworkThe Wiring Diagram.

The computational architecture that mimics the structure of a human brain (loosely). Layers of connected nodes that process information. You will never need to build one. But knowing the term helps when someone at a dinner party tries to explain AI to you incorrectly.

PromptFlight Briefing.

The instructions you give an AI before it does the work. The quality of the briefing determines the quality of the flight. (See: Chapter 2, The Secret Language of the Cockpit.)

Prompt EngineeringFlight Planning.

The skill of writing prompts that get the AI to do what you actually want. Less mystical than it sounds. More important than most people realise.

RAG (Retrieval-Augmented Generation)Pre-Flight Document Pack.

A technique where AI retrieves relevant documents before generating a response, so the answer is grounded in your actual data rather than its general training. NotebookLM uses this. Any AI-powered editorial system can use this.

SaaS (Software as a Service)Leased Aircraft.

Software you access online and pay for monthly rather than buying outright. Claude, Make, Canva, Suno: all SaaS. You don’t own the plane. You rent the seat.

System PromptStanding Orders.

A set of instructions given to an AI at the start of every conversation that shapes its behaviour, personality, and boundaries. Archer’s system prompt is why he sounds like Archer rather than a generic chatbot.

TemperatureThrottle Setting.

A parameter that controls how creative or predictable an AI’s output is. Low temperature = conservative, predictable, factual. High temperature = creative, surprising, occasionally unhinged. Red runs hot. Archer runs cool.

TokenFuel Unit.

The smallest unit of text an AI processes. A word is roughly 1-2 tokens. A page is roughly 500 tokens. When someone says a model has a 200K token context window, they mean it can hold about 400 pages in working memory. That is one example, and a dated one: current windows reach a million tokens.

Training DataFlight Hours.

The information an AI was trained on before you ever talked to it. Determines what it knows, what it doesn’t, and where its biases come from. You can’t change it. You can supplement it with your own documents.

Voice CloningCo-Pilot Impersonation.

AI replicating your voice so convincingly that your own mother might not tell the difference. ElevenLabs does this. The ethical implications are real. The production implications are extraordinary.

Workflow / ScenarioFlight Plan.

A sequence of automated steps that move data between tools and trigger actions. In Make, these are called scenarios. Build one well, and it runs your operation while you sleep.

Zero-Shot / Few-ShotCold Start vs. Pre-Flight Briefing.

Zero-shot: asking AI to do something with no examples. Few-shot: giving it a few examples first so it understands the pattern. Few-shot learning almost always produces better results. Show, don’t just tell.