Reference · Canonical Language

Glossary

The shared, plain-English vocabulary for this course. Every lesson uses these words to mean exactly these things. Keep it nearby.

Talking to AI

AI model (LLM): A program that predicts likely text from the text you give it. Think of it as a fast, confident, well-read intern with no memory of yesterday and no live access to your accounts unless you connect it.
Prompt: The instructions you give an AI model — the request plus any context, examples, and desired format. A good prompt is reusable, not a lucky one-off.
Context: The background information you hand the model inside the prompt (a document, your brand voice, last email) so its answer fits your situation.
Hallucination: When a model states something false with full confidence. The single biggest risk for business use — always verify facts, dates, and numbers.
Token: The unit AI tools count for length and price — roughly ¾ of a word. Matters because longer prompts and outputs cost more and have limits.

Automation: A workflow that runs on its own: "when X happens, do Y" — without you clicking anything each time.
Trigger: The event that starts an automation — a new email, a form submission, a new row in a sheet. The "when X happens" half.
Action: What the automation does in response to a trigger — send a reply, add a row, summarize text with AI. The "do Y" half.
No-code tool: A platform (e.g. Zapier, Make) where you build automations by connecting apps visually, with no programming.
Human-in-the-loop: A checkpoint where a person reviews or approves before the automation acts — your safety valve for anything customer-facing or risky.
Agent: An AI setup that decides which steps to take to reach a goal, rather than following a fixed "when X do Y" path. More powerful, harder to control — use only when a fixed automation can't do the job.

Vibe coding: Building software by describing what you want to an AI and iterating on what it makes — instead of writing the code yourself. You steer; the AI types.
Build brief: A short, plain-English spec of what you're building: who it's for, the one job it does, its inputs and outputs, and what "done" means. The biggest lever on output quality.
Code: The text instructions that make an app work. You don't need to write it from scratch — but you'll learn to read it well enough to find and fix things.
Debugging: Finding and fixing what's broken. With AI: copy the error message, give it the context, isolate the one thing that changed, fix, test. A loop, not a guess.
Git / repository: A save-points-and-undo-history system for your project. The repository ("repo") is the project's folder of code plus its full history, so you can always roll back to a version that worked.
Deploy: Putting your app on the live internet so others can use it at a real web address (URL). "Shipping" it.
Comprehension debt: The trap of owning an AI-built app you can't read, trace, or fix — so every problem means burning credits guessing. The thing this course's build half is designed to keep you out of.

Tool: A single defined ability you grant an agent so it can act — search the web, send an email, query a database. The agent can only do what you've unlocked.
CLI (command line): Controlling software by typing commands instead of clicking. Giving an agent a terminal lets it use the whole universe of existing command-line tools with no special integration — how coding agents like Claude Code work.
MCP (Model Context Protocol): An open standard — "USB for AI" — that lets any agent plug into any tool or data source through a ready-made, permissioned connector instead of custom wiring. More structured and reusable than a raw CLI; only works where an MCP server exists.
Skill: A bundle of instructions (a SKILL.md file) that teaches an agent how to do something well — expertise it loads on demand. The teach skill installed at the start of this course is one.