The AI Agent Glossary (Plain English)

You use AI every day, but half the words around it — harness, orchestrator, MCP — sound like jargon built to keep you out. Here are the ten terms that actually matter, each in one plain sentence, plus why you should care.

One picture to hang it all on: a restaurant kitchen. The AI model is the chef. Everything below is the kitchen around that chef — the recipes, the runners, and the head chef keeping it all moving.

The Terms

Agent — An AI that doesn't just answer, it acts: it takes a goal, decides the steps, uses tools, and keeps working until the job is done. Why it matters: this is the jump from AI that talks to AI that ships — the whole point of climbing the Levels of the Game.

Agent harness — The software wrapped around the model that runs the actual work: it reads the model's output, runs the tools it asks for, feeds the results back in, retries failures, and decides when to stop. Why it matters: the harness is the difference between a chef describing a dish and a working kitchen that plates it.

Agent orchestrator — A layer above the harness that coordinates several agents at once, handing tasks between them (frameworks like LangGraph and CrewAI do this). Why it matters: it's the head chef running multiple stations — you rarely touch it early, but it's how big jobs get split up.

Context Files — Plain markdown files you write and hand your AI that tell it who it is, what the rules are, and how you want the work done. Why it matters: this is the highest-leverage thing you control — good Context Files turn a generic chef into your chef.

MCP (Model Context Protocol) — An open standard from Anthropic (introduced late 2024, now widely adopted) that lets an AI plug into outside tools and data — your email, your files, your calendar — through one common connector. Why it matters: it's the universal outlet, so you stop copy-pasting between apps and let the AI reach them directly.

Devcontainer — A pre-packaged, self-contained workspace that gives the AI a clean, consistent place to run with the right tools already installed. Why it matters: "it works on my machine" stops being a problem — the kitchen comes fully stocked every time.

Agentic loop — The cycle an agent repeats: think, act (use a tool), read the result, decide the next step — over and over until the job is done. Why it matters: once you see the loop, you stop expecting one perfect answer and start trusting the AI to work in passes.

Context engineering — The craft of deciding what goes into the AI's limited working memory each turn, and what to leave out. Why it matters: it's mise en place — a prepped, focused station beats a counter buried in clutter, and it's what separates sloppy results from sharp ones.

Skills — Reusable, packaged instructions (often a folder of markdown and files) that teach an agent one task well, loaded only when it's needed. Why it matters: skills are recipe cards pulled off the shelf on demand — build a task once, reuse it forever.

Tool call — The moment the AI stops writing text and actually does something: searches the web, sends an email, edits a file. Why it matters: tool calls are where talk becomes action, and action is where results come from.

How this connects to the Engine

Nearly every term here loops back to the one lever you actually control: your Context Files. The harness, the tools, the loop — those run in the background, but the markdown you hand your AI is what makes the output yours. If writing those files from a blank page feels daunting, the $1 Starter Kit generates your first set for you, so you start with something real to edit instead of a blinking cursor.