Skip to content

Glossary

Plain-language definitions for the terms you'll meet on the ramp. Unofficial — when in doubt, Microsoft's Copilot hub is the source of truth.

How to use this page

Skim the group that matches where you are on the journey. Every term links to the stage or walkthrough where it actually shows up, so you can go from "what is that?" to "show me" in one click.

Borrow, don't build

These are plain-language definitions to get you oriented — Microsoft owns the authoritative, always-current glossary for each product. When a definition matters for customer-facing work, go to the source.

Core products

  • Microsoft 365 Copilot — the paid Copilot that's grounded in your work content (email, chat, files, meetings) through the Microsoft Graph. It's what powers most of Stage 1 through Stage 3. Requires a per-user license on top of a qualifying M365 plan.
  • Copilot Chat — the chat surface (in Teams, Outlook, the M365 app, and the web). With a Microsoft 365 Copilot license it can reason over your work content; without one it still answers general web-grounded questions. The home of Stage 1.
  • BizChat — the "chat over all your org content" experience inside Microsoft 365 Copilot — it searches across SharePoint, email, and Teams and returns a single cited answer.
  • Copilot Studio — the low-code platform for building production agents: custom topics, connectors, authentication, publishing, and governance. The low-code maker destination, Stage 6.
  • Agent Builder — the lightweight, in-product agent maker inside Microsoft 365 Copilot. You point it at knowledge, give it instructions, and share it — no code. Your first taste of making, Stage 4.
  • Microsoft Foundry (formerly Azure AI Foundry) — the pro-code platform for building custom AI apps and agents with full control over models, orchestration, and evaluation. The pro-code frontier, Stage 7 — one rung beyond Studio for advanced builders.

Journey concepts

  • First-party agents — the prebuilt agents Microsoft bundles into the Microsoft 365 Copilot license (Researcher, Analyst, Facilitator, Interpreter, Project Manager, and more). You delegate to them rather than building anything. The whole of Stage 2.
  • Cowork — Microsoft 365 Copilot's ability to take a whole multi-step task — plan it, run the steps across your tools, and hand back a finished result — instead of answering one prompt at a time. The focus of Stage 3.
  • Copilot Pages — a shared, editable canvas. Turn any Copilot response into a living document your team can co-edit with Copilot in the loop.
  • Frontier — Microsoft's early-access program. The newest Copilot capabilities (including the fuller Cowork feature set) often land in Frontier first, weeks or months ahead of general availability. See the Cowork stage for how to check eligibility and enroll.
  • Autopilot (category) — Microsoft's category of always-on, autonomous agents: each runs under its own governed Microsoft Entra identity and acts on your behalf in the background, within the permissions and policies you set, instead of waiting for a prompt. Announced June 2, 2026. The focus of Stage 5.
  • Microsoft Scout — the first Autopilot: an always-on personal agent that builds context over time through Work IQ and can coordinate meetings, prep, and deliverables for you. In Frontier private preview today. (Microsoft Scout on Learn)
  • Declarative agent — an agent defined by configuration, not code: instructions, knowledge sources, and starter prompts. What you build in Agent Builder, and the simpler option in Studio.
  • Custom engine agent — an agent with its own orchestration logic and (optionally) its own model, built in Copilot Studio or Microsoft Foundry. More power, more responsibility — for when a declarative agent isn't enough.

  • Harness (Studio engine) — the engine underneath a Copilot Studio agent: it sets how autonomously the agent reasons, what files it can touch, where you can publish it, and — critically — how it bills and whether your existing licenses cover it. Same builder, three engines today:

    • GitHub Copilot harness — autonomous and agentic: reasons through a goal step by step, retries, works with Word/Excel/PowerPoint/PDF files, and has skills and memory. The modern direction for complex, multi-step work. Bills Copilot Credits for all usage — creation and runtime — and is never covered by a Microsoft 365 Copilot license.
    • Standard harness — follows the topics and rules you author: predictable, structured conversations. Usage is covered by a Microsoft 365 Copilot license when it runs inside Microsoft 365 channels (billed only for unlicensed users or use outside those channels).
    • Copilot chat harness — extends Microsoft 365 Copilot Chat with your enterprise knowledge; included in Microsoft 365 Copilot USLs / consumption.

    Pick for the job, then confirm the cost and coverage implications — see Pick the engine for the job and How Copilot Credits work.

Building blocks

  • Skill — a reusable, packaged capability you save once and reuse or share. Today the clearest example is a Cowork recipe (Stage 3). In Autopilots — Microsoft's category of always-on agents — that same instinct is equipped onto Microsoft Scout, the first Autopilot, so it draws on the capability in the background (Frontier private preview). In Copilot Studio, skill is a precise term for one specific kind of tool — an embedded pro-code agent (Bot Framework / Microsoft 365 Agents SDK) registered so your agent can call it; the connector actions, flows, and MCP tools in Stage 6 are tools, not skills. See the Skills Catalog. Not the first-party Skills agent (career stretch assignments) or your personal skill level.
  • Grounding — giving an agent the specific content it should answer from (a SharePoint site, a set of files, a database) so its answers are based on your facts, not the model's general training.
  • Knowledge source — the documents or sites an agent is grounded on. The single biggest driver of answer quality.
  • Topic — in Copilot Studio, a defined conversation path that handles a particular intent (e.g. "reset my password"). The building block of structured, multi-turn conversations.
  • Connector / action — a configured capability that lets an agent do something in another system — look up an order, create a ticket, update a record — rather than only talking. See Connect an agent to a system.
  • Power Automate flow — a workflow that an agent can trigger to act on business systems (create a ticket, send a notification, write to a database). See Trigger a Power Automate flow.
  • MCP (Model Context Protocol) — an open standard for connecting agents to external tools and data sources in a consistent way. See Add an MCP / tool integration.
  • Autonomous / triggered agent — an agent that acts when an event happens (a file is added, a message arrives) instead of waiting for someone to start a conversation.

Cost & consumption

  • Copilot Credit (formerly "message") — the unit Copilot Studio meters agent usage in. Credits are pooled across the whole tenant and funded through pay-as-you-go, prepurchase, or prepaid packs. The name changed from messages on September 1, 2025 (same quantity, same rate — just a new name). Learn the model on How Copilot Credits work and estimate yours in the Credit Estimator.
  • Classic vs generative answer — a classic answer is a pre-authored, static reply (1 credit); a generative answer has the AI read your knowledge and respond dynamically (2 credits). Most modern agents are generative.
  • Agent action — a step where the agent does something — a connector or flow call, deep reasoning, or a topic transition — billed at 5 credits.
  • Tenant graph grounding — retrieval-augmented generation over your tenant-wide Microsoft 365 Graph (people, mail, meetings, files), billed at 10 credits on top of the answer. Optional and toggled per agent — the powerful-but-pricier grounding mode, distinct from free grounding over your own documents.
  • Interactive vs autonomous (billing regimes) — interactive agents bill per user interaction; on the standard harness those interactions are zero-rated for Microsoft 365 Copilot-licensed users inside Microsoft 365 channels. Autonomous / triggered agents bill per event, with no license discount. This regime interacts with your harness choice: the GitHub Copilot harness is never license-covered — every interaction consumes Copilot Credits regardless of license. Which regime and harness an agent uses usually dominates its bill.
  • Build size (T-shirt) — an XS → XL estimate of the effort to build and own an agent — not its run cost. A cheap-to-run agent can be expensive to build, and a tiny agent at high volume can be expensive to run; the estimator reports the two axes separately.

Platform & governance

  • Power Platform — Microsoft's low-code suite (Power Apps, Power Automate, Power Pages, and Copilot Studio). Studio agents live in a Power Platform environment.
  • Environment — an isolated container in Power Platform where agents, flows, and data live. Used to separate dev, test, and production.
  • Microsoft Entra ID (formerly Azure AD) — Microsoft's identity service. It's how agents authenticate users and act with the right permissions. See Secure a Studio agent.
  • ALM (Application Lifecycle Management) — the practice of moving an agent through dev → test → production in a controlled, repeatable way, with versioning and governance.
  • DLP (Data Loss Prevention) — admin-defined policies that control which connectors and data an agent is allowed to touch — a core governance guardrail.
  • Tenant — your organization's instance of Microsoft 365. Licensing, admin settings, and feature availability (including Frontier) are all set at the tenant level. Copilot Studio usage is metered in Copilot Credits — see Cost & consumption above.

📚 Learn more.

Was this page helpful?

Found an issue or want to suggest an improvement? Open a GitHub issue  ·  Star on GitHub