Books · AI-Powered Productivity for Microsoft Office Professionals · free chapter

Part 1 · Chapter 03 ≈ reading…

Chapter 3: AI Jargon, Explained

Plugins, skills, hooks, agents, MCP and the rest, in plain language

"You do not need to build AI to use it well. You do need to know what the words mean, because the words tell you what the tool can reach and what it is allowed to do."

In this chapter

Part 1 · Foundations

Every AI word, explained like you would explain it to a new colleague.

AgentsSkillsPluginsMCPHooks

What You'll Learn

The basics every Office professional hears: model, prompt, grounding, context and hallucination
The building blocks of AI help: agents, instructions, knowledge, skills, tools and actions
The plumbing that connects AI to your work: connectors, plugins, MCP and hooks
How the pieces fit together, using one picture and one analogy
Which words are Microsoft's and which come from the wider AI tool world

Why the Words Matter

AI vocabulary moves fast, and vendors do not always use the same word for the same thing. That would be a curiosity if the words were only marketing. They are not. Each term tells you something practical: where the AI gets its information, what it is allowed to do, and who is responsible when it acts.

This chapter is a reference you can come back to. Every term gets a plain language definition, an Office example, and a place in one picture at the end.

One Analogy for Everything

Imagine your team hires a brilliant new colleague. They are well read, fast and eager, but they know nothing about your company yet. Every AI term maps to something you would do to get that person productive.

The new colleague analogy

Each AI term, mapped to onboarding a capable new hire

AI term	The new colleague	In Office
Model	Their education and general knowledge	The large language model behind Copilot
Prompt	What you ask them to do	"Summarize this thread in five bullets"
Grounding	The files you hand them before they start	Pointing Copilot to a workbook or a document
Instructions	Their job description	The standing instructions inside an agent
Skill	A playbook for one recurring task	Excel's `@check-model` review routine
Tool or action	A system they are allowed to use	Creating a file, sending a draft, querying data
Connector	A keycard to another building's records room	Bringing CRM or ticketing data into Copilot
MCP	A universal plug that fits every socket	One standard way for AI to call any tool
Plugin	The onboarding kit that bundles the playbooks and keycards	A package of skills and tools you add in one step
Agent	The colleague working through a task on their own	Agent mode, Researcher, a custom agent
Hook	A company rule that runs every time, no exceptions	A check that runs automatically before an AI tool acts

Figure 3.1. If a term confuses you, ask what it would mean for a new hire. The analogy is rarely perfect and almost always useful.

The Glossary

Filter the cards by group, or read them all. Microsoft specific terms are marked with the Microsoft 365 logo.

AI terms for business professionals

Plain language first, then an example from Office

Model

The large language model (LLM) that reads text and predicts a useful response. It knows a lot in general and nothing about your company until you give it context.

Office example: Copilot uses models such as OpenAI's GPT family and Anthropic's Claude behind the scenes.

Prompt

What you type or say to the AI. A good prompt names the goal, the source, the reader and the format.

Office example: "Draft a one page proposal from /Q3 plan.docx for the finance team, formal tone."

Grounding

Giving the AI real sources to answer from, so it works from your facts instead of its general memory.

Office example: Referencing a file with / in the Copilot prompt box.

Context window

How much text the AI can hold in mind at once: your prompt, the files and the conversation so far. Very long material may be summarized or cut.

Office example: Why a focused question about one section beats "read all 400 pages".

Hallucination

A confident answer that is wrong or made up. Grounding reduces it; reviewing catches it.

Office example: A summary that quotes a figure the source never contained.

Copilot

Microsoft's name for its AI assistant. It appears in Word, Excel, PowerPoint, Outlook and Teams, and as a chat app.

Office example: The Copilot button on the Home tab in Word.

Agent

AI that can plan and take several steps toward a goal, using knowledge, skills and tools, instead of answering one prompt at a time.¹

Office example: Edit with Copilot in Excel building a model step by step; the Researcher and Analyst agents.

Instructions

Standing guidance an agent always follows: its role, tone, rules and limits. You write them once and they apply to every conversation.

Office example: A custom agent told to answer only from the HR policy site.

Knowledge

The sources an agent is allowed to draw on, such as SharePoint sites, files or web pages.

Office example: An agent grounded on the sales playbook library.

Skill

Reusable instructions or a workflow for one specific job: the steps, checks and output an expert would use.³ You call it by name instead of writing the whole prompt each time.

Office example: Typing @ in Copilot in Excel and choosing a built in skill such as @check-model.

Tool or action

Something the AI can do, not just say: create a file, run a query, send a draft, update a record. Actions are where permissions and approvals matter most.

Office example: An agent that files a ticket in your service desk system.

Connector

A Copilot connector (formerly a Graph connector) brings content from another system into Microsoft 365, so Copilot can reason over it alongside your files and mail.⁴

Office example: One of the 100+ prebuilt connectors indexing your ticketing system.

Plugin

An installable package that brings together what Copilot needs for one kind of task: an agent, skills, connectors and MCP integrations.² Older documents use "plugin" for a set of actions that call an API.

Office example: The legal plugin in Word, managed under Manage plugins & skills.

MCP

Model Context Protocol: an open standard for connecting AI to tools and data. Build a tool once as an MCP server and many AI apps can use it, including Copilot Studio agents.⁵

Office example: A Copilot Studio agent calling a company system through an MCP server.

Hook

An automatic step that runs every time a set event happens, such as a session starting or a tool about to run. A hook can check, log or block the action. Builders set hooks; end users rarely see them.⁶

Example: Copilot Studio hooks (preview) run a workflow before an agent uses a tool. Claude Code and GitHub Copilot use hooks the same way.

Microsoft Graph

The map of your organization's work: people, files, mail, meetings, chats and how they relate. Copilot uses it to find what is relevant to you.

Office example: Why Copilot can answer "what did Ana send me about the budget?"

Frontier

Microsoft's program for early access to preview AI features before they are generally available. Your organization opts in; features can change before release.⁷

Office example: Trying a new agent before it rolls out to everyone.

Figure 3.2. Sixteen terms in four groups. The plumbing group is where Office and the wider world of AI builder tools meet.

Skills, Plugins and Hooks, Up Close

Three words cause the most confusion because they are new to most business professionals and they overlap. Here is the short version.

A skill is know-how. It is a written routine for one task: what to check, in what order, and what the result should look like. In Excel and PowerPoint you call a skill with @ or from the + menu. Under the hood a skill is a folder with a SKILL.md file, an open standard that Claude Code and GitHub Copilot use too, so you can write your own and keep them in OneDrive.³
A plugin is a package. It bundles an agent, skills, connectors and MCP integrations so you can add them in one step and keep them up to date. Think of it as an app store install for AI capabilities.
A hook is a guardrail. It runs by itself every time an event happens and can stop an action. A tool runs when the agent decides it is relevant; a hook runs every time, no matter what.⁶ Skills make AI more capable; hooks make it more predictable.

How the Pieces Fit

Put the terms together and a simple picture appears. You ask; the AI app decides what it needs; it uses its skills and tools to reach your data; and everything it does stays inside your permissions.

One request, every layer

"Check my budget model and summarize the risks for the CFO"

YouPrompt

The goal, the file and the reader "for the CFO"

The AI appCopilot or an agent

Copilot in Excelplans the stepsuses a model

Know-howSkills and instructions

@check-modelyour team's review rules

ReachTools, connectors, MCP

read the workbooksearch your filescall a company system

BoundariesPermissions and rules

only what you can accessadmin policieshooks set by builders

Figure 3.3. Five layers. When an answer surprises you, ask which layer caused it: a vague prompt, a missing skill, a source it could not reach, or a permission it did not have.

Try It Yourself

Three exercises that turn vocabulary into instinct.

Find the skills menu. Open Copilot in Excel, type @ in the prompt box and read the list of skills you are offered. Then open the + menu and look for plugins and skills.
What good looks like
You can name two skills available to you and say in one sentence what each does. If you see none, check that your Office display language is English, since some skills require it, and ask your administrator whether they are enabled.
Ground the same question twice. Ask Copilot Chat a question about a project without referencing any file. Then ask again and reference the project document with /.
What good looks like
The grounded answer cites your document and is specific; the first answer is generic. That difference is grounding.
Explain one term with the analogy. Pick MCP, plugin or hook and explain it to a colleague using the new colleague analogy, in two sentences.
What good looks like
Your colleague can repeat it back in their own words. If they cannot, the analogy needs one more concrete example from your work.

Key Takeaways

Words tell you reach and responsibility: where AI gets its facts, what it can do, and who answers for it.
Grounding gives AI your facts; reviewing catches what it still gets wrong.
Agents take several steps with tools; skills give them know-how for one task.
Plugins package capabilities; connectors and MCP connect AI to systems.
Hooks are automatic guardrails that builders set in tools like Copilot Studio; they make AI predictable.
Permissions sit under every layer: Copilot only reaches what you can already reach.

What's Next

Part 2 puts the vocabulary to work, app by app, starting with Word: drafting, rewriting, reviewing and, for legal teams, a set of purpose built skills.

References

Microsoft 365 Copilot extensibility overview. Microsoft Learn, accessed October 2026. learn.microsoft.com
Plugins for Microsoft 365 Copilot. Microsoft Learn, accessed October 2026. learn.microsoft.com
Add skills to declarative agents. Microsoft Learn, accessed October 2026. learn.microsoft.com
Microsoft 365 Copilot connectors overview. Microsoft Learn, accessed October 2026. learn.microsoft.com
Connect your agent to an existing Model Context Protocol (MCP) server. Microsoft Copilot Studio, Microsoft Learn, accessed October 2026. learn.microsoft.com
Hooks overview (preview). Microsoft Copilot Studio, Microsoft Learn, accessed October 2026. learn.microsoft.com. See also Hooks reference, Claude Code docs. code.claude.com
Microsoft 365 Copilot Frontier program. Microsoft, accessed October 2026. www.microsoft.com

Microsoft, Microsoft 365, Copilot, Word, Excel, PowerPoint, Outlook and Teams are trademarks of the Microsoft group of companies. Product icons are shown for training and documentation. This book is independent and is not affiliated with, sponsored by or endorsed by Microsoft. Product details and prices were checked in October 2026 and change often.