AI terms and file types a non-technical person should know

You need about twenty words to follow almost any conversation about AI. This glossary covers the terms and file types you’ll run into when you use AI assistants, grouped so you can read it top to bottom or jump to what you need. Each entry is a sentence or two, with an example where it helps.

Talking to AI

Prompt. What you type to the AI. A question, an instruction, a pasted document with a request. “Summarise this email in three bullets” is a prompt.

System prompt. Instructions the AI gets before your conversation starts, usually set by whoever built the app. It’s why the same model behaves differently in different products.

Chat / conversation. One continuous exchange. The AI can see earlier messages in the same chat, and usually nothing from your other chats.

Context. Everything the AI can see right now: your messages, its replies, any files you attached. Better context gives better answers, which is why people talk about “context engineering”.

Hallucination. When an AI states something false with total confidence, such as a made-up quote or a wrong date. It isn’t lying on purpose. It’s producing what sounds right.

Under the hood

Model. The trained AI itself. Claude, GPT and Gemini are families of models. Bigger models are usually smarter and slower. Smaller ones are faster and cheaper.

LLM (large language model). The kind of model behind chat assistants. It learned from a huge amount of text and works by predicting what comes next. I explain this in what is AI, in simple terms.

Token. The unit models read and write in. A token is a word or a piece of a word, roughly three-quarters of an English word on average. Limits and prices are counted in tokens.

Context window. The most text a model can keep in view at once, measured in tokens. If a conversation or document is longer than the window, the earliest parts fall out of view.

Training / training data. How a model learns, and what it learned from. Training happens before you ever use it. Chatting with a model doesn’t retrain it on the spot.

Knowledge cutoff. The date a model’s training data ends. Without web search, it won’t know about anything after that.

AI that does things

Agent. An AI that takes actions, not just answers. Instead of telling you how to rename fifty files, an agent renames them. It works in a loop: decide a step, do it, look at the result, decide the next step.

Tool. Something an agent is allowed to use: web search, a calculator, a file editor, your calendar.

MCP (Model Context Protocol). A standard way to plug tools and apps into an AI assistant, a bit like USB for AI. If an app “has an MCP server”, an assistant can connect to it.

Skill. A saved recipe for a task, usually a short written guide plus a script. I use one to release my projects: I say “ship this”, and the assistant follows the recipe step by step. I describe it in shipping solo with AI.

Vibe coding. Building software by describing what you want and letting AI write the code. More in what is vibe coding.

File types you’ll meet

You’ll see these because AI reads and writes them constantly. All of them are plain text. You can open any of them in a basic text editor.

File	Ends in	What it’s for
Plain text	.txt	Text with no formatting at all
Markdown	.md	Text with light formatting, like headings and lists. The format AI writes in
HTML	.html	A web page. Double-click it and it opens in your browser
CSS	.css	How a web page looks: colours, fonts, spacing
JavaScript	.js	What a web page does: buttons, timers, games
JSON	.json	Structured data: settings, lists, records
CSV	.csv	A simple table. Opens in Excel or Google Sheets
PDF	.pdf	A fixed document for reading and printing. Not plain text

Markdown deserves its own entry because it’s everywhere in AI. When an assistant uses **bold** or starts a line with #, that’s Markdown. I wrote a short guide: Markdown explained.

HTML is the beginner’s superpower. Ask an AI for “a single HTML file” and you get a working app you can open with a double-click, with nothing to install. One of my tools, Benztormr, is exactly that: one HTML file.

JSON looks technical and isn’t. It’s labels and values:

{
  "title": "Kaiju City Smash",
  "players": 4,
  "online": true
}

One of my projects turns a single JSON file like this into a finished video.

README. A file, usually README.md, that explains what a project is and how to use it. It’s the first thing to read in any project folder.

CLAUDE.md / AGENTS.md. A notes file that an AI coding assistant reads before working on a project. It holds the rules and preferences for that project. Here’s how to write one.

Building and sharing

Repository (repo). A project folder that remembers every saved version of itself.

Git and GitHub. Git is the tool that saves the versions. GitHub is the website that stores them online. Here’s the difference, and here’s why I think everyone should learn it.

Commit. One saved version of a project, with a note about what changed.

API. A way for one program to talk to another. When an app “uses the Claude API”, it’s sending your text to the model and getting the answer back, with no chat window involved.

API key. A password for an API. Treat it like a bank PIN. Never paste one into a chat, a screenshot or a public file.

Deploy. Put something on the internet so other people can use it.

Domain. A web address, like karankharode.com.

Open source. Software whose code is public, so anyone can read it, use it and improve it.

How to use this list

You don’t need to memorise it. When you hit an unfamiliar word, the fastest fix is to ask the AI itself: “Explain what a context window is, like I’m new to this, with one example.” Then come back here if you want the short version.

If you’re just getting started, read these three next: what AI is, Markdown explained and everyone should learn GitHub.