Microsoft 365 Copilot is the AI assistant built into Microsoft 365 apps that uses large language models and Microsoft Graph data to help users draft, summarize, and analyze content.
Microsoft 365 Copilot combines large language models with Microsoft Graph — the data layer connecting a user’s emails, calendar, chats, files, and meetings — to generate contextually relevant responses inside apps such as Word, Excel, PowerPoint, Outlook, and Teams. Because Graph enforces existing Microsoft 365 permissions, Copilot never surfaces content the user cannot already access, making data governance a central admin concern. For the AB-900 exam, distinguish Microsoft 365 Copilot (the licensed, app-embedded assistant requiring a base license plus a paid per-user add-on) from the free consumer Microsoft Copilot chat experience, which differs in licensing, data access, and admin controls.
An agent is an AI-powered assistant that automates or executes business processes, ranging from simple prompt-and-response helpers to fully autonomous workers that act alongside or on behalf of people.
In Microsoft 365 Copilot, an agent is an AI-powered assistant scoped to specific instructions, data sources, and actions rather than general productivity assistance. Agents are built in Copilot Studio and deployed within Teams, SharePoint, or other Microsoft 365 surfaces, ranging from simple prompt-guided helpers to more autonomous workflows that take multi-step actions. A key exam distinction is that an agent is not Copilot itself: Copilot offers broad, app-embedded help, while an agent answers only from the knowledge and tools its creator configured. Admins govern agent availability through the Integrated Apps section of the Microsoft 365 admin center, part of the Copilot Control System.
Researcher is a first-party reasoning agent in Microsoft 365 Copilot that performs complex, multi-step research across work data and the web to produce comprehensive, source-cited reports.
Researcher is a built-in, first-party agent in Microsoft 365 Copilot that pairs a deep-research reasoning model with Copilot orchestration to run multi-step investigations across work data — emails, meetings, files, and chats — and the web, then returns a synthesized, source-cited report. As a first-party surface it is governed through Copilot licensing and access rather than deployed like a custom Copilot Studio agent. The exam contrast is with Analyst: Researcher does qualitative, multi-source synthesis and decision-ready summaries, while Analyst handles quantitative reasoning over structured data. Both appear under Tools in Copilot Chat.
Analyst is a first-party reasoning agent in Microsoft 365 Copilot that performs advanced data analysis, using chain-of-thought reasoning and running Python to answer complex data queries.
Analyst is a first-party reasoning agent in Microsoft 365 Copilot built for multi-step quantitative work. Given data such as a spreadsheet, it applies chain-of-thought reasoning and writes executable Python to clean, transform, and summarize the data, showing its code in real time so users can inspect the logic. The key exam distinction is that Analyst targets structured data-analysis tasks while Researcher targets broad information synthesis across web and organizational sources. Both are reasoning agents in Copilot Chat requiring a Microsoft 365 Copilot license, so picking the wrong one for a data scenario is a common distractor.
A custom agent is an organization-built agent tailored to specific tasks, instructions, knowledge sources, and actions for a team or business process.
A custom agent is an organization-authored AI assistant built for a specific workflow, such as HR onboarding or IT helpdesk triage. It is configured with tailored instructions, grounded in company knowledge sources (SharePoint sites, files, or connectors), and optionally extended with actions that call external APIs or Power Automate flows. Custom agents are created in Microsoft Copilot Studio or the simpler Agent Builder within Microsoft 365 Copilot Chat. For the exam, contrast it with a declarative agent, which reuses Microsoft 365 Copilot’s own model and orchestrator and needs a Copilot license; a Copilot Studio custom engine agent uses its own orchestration and can serve users without one.
Microsoft Copilot Studio is the full-featured, low-code tool for building, customizing, and managing agents with advanced tools, connectors, and topics.
Microsoft Copilot Studio is a standalone, low-code platform built on Power Platform for creating custom agents using topics, triggers, entities, actions, and connectors. Unlike the lightweight Agent Builder embedded in Microsoft 365 Copilot, it supports multi-turn conversations, external connectors, Power Automate flows, and publishing to channels like Microsoft Teams and Direct Line, so an agent can reach beyond Microsoft 365 data. For AB-900, the key administrative point is that Copilot Studio usage is measured by consumption (capacity/credits, not per-user), and both capacity allocation and environment assignment are governed in the Power Platform admin center.
The Microsoft 365 Copilot Agent Builder is a simple, in-context experience within Copilot Chat for quickly creating agents without leaving the chat interface.
Agent Builder is the lightweight, in-context authoring experience embedded in Microsoft 365 Copilot Chat, letting users create a custom agent by describing its purpose, adding knowledge such as SharePoint files or websites, and selecting capabilities — all without leaving the chat. Requiring no separate app or developer tooling, it suits business users, not just IT. The exam distinction is the capability ceiling: Agent Builder handles simple, narrowly scoped conversational agents, while Copilot Studio supports complex logic and broader governance. Admins manage availability through Microsoft 365 admin center Copilot settings.
A declarative agent is an agent defined by instructions, actions, and knowledge sources on top of the Microsoft 365 Copilot orchestrator and foundation models, rather than custom-built infrastructure.
A declarative agent extends Microsoft 365 Copilot by supplying natural-language instructions, optional knowledge sources (SharePoint sites, public URLs, or uploaded files), and actions (API plugins or connectors) without writing orchestration code. The Copilot orchestrator handles grounding, reasoning, and response generation using the underlying foundation models; the author simply declares what the agent knows and can do. Contrast this with a custom engine agent, which brings its own language model and orchestration for full control. Agents grounded only in instructions and public web content carry no extra charge, while those reaching tenant data trigger metered, pay-as-you-go billing.
Prompt management covers saving, sharing, scheduling, and deleting reusable prompts, primarily through Copilot Prompt Gallery.
Prompt management covers the lifecycle of reusable prompts in Microsoft 365 Copilot: creating, saving, liking, sharing, and deleting them through the Copilot Prompt Gallery, which surfaces across apps like Teams, Word, and Copilot chat. Users can keep prompts private or share them with a team for consistent phrasing. The exam-critical admin angle is that an account with the Global Administrator role can export a user’s saved, liked, and shared prompt data using a PowerShell cmdlet, a data-governance task distinct from the usage and adoption metrics shown in Copilot Analytics.