Skip to content

Annex A — Glossary

Consolidated definitions. Every 📖 Definition callout in the book feeds this glossary — one wording per term.

  • Agent — A purpose-built assistant on Microsoft 365 Copilot, scoped to a specific job with its own instructions, knowledge, and suggested prompts; reusable and shareable. (Ch 8)
  • Agent Store — The catalog for finding and adding ready-made agents from Microsoft, partners, and your organization, without building anything. (Ch 8)
  • Analyst (in Copilot) — A built-in Microsoft 365 Copilot agent for quantitative data analysis — works through raw data to produce insights, tables, and visualizations. (Ch 11)
  • Artificial intelligence (AI) — Software that performs tasks normally associated with human intelligence, such as recognizing images, understanding language, or making recommendations. (Ch 1)
  • Azure AI Search — A Foundry tool providing retrieval/grounding (RAG) over your data for custom AI apps; the custom-build counterpart to Copilot's semantic index. (Ch 13)
  • Azure AI Vision (in Foundry Tools) — A Foundry capability for image understanding, such as extracting text from scanned documents. (Ch 13)
  • Bias — Systematic skew in output caused by unrepresentative training data or design, producing unfair or inaccurate results. (Ch 4)
  • Copilot agent builder — The no-code tool for creating a declarative agent by describing its behavior, giving it knowledge, and configuring settings. (Ch 8)
  • Copilot Pages — A persistent, collaborative canvas that turns Copilot responses into durable content colleagues can edit together. (Ch 10)
  • Copilot Prompt Gallery — The library in Microsoft 365 Copilot to discover, save, edit, and share prompts. (Ch 6)
  • Fabrication (hallucination) — Confident, plausible-sounding content that is factually wrong or invented; a consequence of predicting plausible text. (Ch 1, 4)
  • Fine-tuned model — A pretrained model further trained on a smaller, domain-specific dataset to perform better on a specialized task. (Ch 1)
  • Generative AI — A category of machine learning that generates new original content — text, images, code, audio, video — in response to a prompt. (Ch 1)
  • Grounding — Providing input sources related to a prompt so responses are more accurate and relevant; in Microsoft 365 Copilot, from Microsoft Graph and optionally the web. (Ch 2)
  • Instructions — Standing guidance you give Copilot (tone, format, role) applied across interactions; also an agent-configuration element. (Ch 8, 10)
  • Large language model (LLM) — An AI model trained on large amounts of text that predicts the next token to generate, summarize, translate, or classify text. (Ch 1)
  • Machine learning (ML) — The branch of AI where a model learns patterns from example data instead of being explicitly programmed. (Ch 1)
  • Management summary — A concise, executive-oriented overview of a longer document — key points, decisions, and implications. (Ch 9)
  • Memory — Copilot's ability to remember useful context about you to make responses more relevant over time. (Ch 10)
  • Microsoft 365 Copilot — The AI assistant embedded across Microsoft 365, grounded in your work data via Microsoft Graph; licensed per user. (Ch 2, 11)
  • Microsoft Copilot (Copilot Chat) — The general AI chat experience on web and mobile; can use work data with a license. (Ch 11)
  • Microsoft Copilot Studio — A low-code platform to build, customize, and manage agents and extend Microsoft 365 Copilot. (Ch 12)
  • Microsoft Foundry (Azure AI Foundry) — Microsoft's platform for designing, customizing, and managing custom AI applications and agents, with a model catalog and lifecycle tools. (Ch 13)
  • Microsoft Graph — The gateway to your organization's Microsoft 365 data (emails, chats, meetings, files) and relationships; the grounding source for Copilot. (Ch 2, 12)
  • Microsoft 365 Copilot connector — A way to bring external data into Microsoft Graph so Copilot can ground on it, respecting permissions. (Ch 12)
  • Notebook — A personal Microsoft 365 Copilot workspace to gather content and prompts on a topic and keep working over time. (Ch 7)
  • Over-reliance — The human tendency to accept AI output uncritically, without appropriate verification. (Ch 4)
  • Pretrained model — A model already trained on a broad, general-purpose dataset that works out of the box for many tasks. (Ch 1)
  • Prompt — The natural-language input (instruction, question, context) you give the model to steer its output. (Ch 1, 3)
  • Prompt engineering — Designing and refining prompts (instructions, context, sources) to get more accurate, relevant, useful output. (Ch 3)
  • Prompt injection — An attack where malicious instructions hidden in content try to hijack the model into ignoring its rules or leaking data. (Ch 4)
  • Researcher (in Copilot) — A built-in Microsoft 365 Copilot agent for complex, multi-step research over work data and the web, producing thorough, cited outputs. (Ch 11)
  • Responsible AI — Developing, deploying, and using AI safely, trustworthily, and ethically, guided by defined principles (fairness, reliability & safety, privacy & security, inclusiveness, transparency, accountability). (Ch 4, 15)
  • Retrieval-augmented generation (RAG) — An AI pattern that retrieves relevant data and adds it to the prompt before the model generates an answer; Copilot's grounding is an enterprise RAG implementation. (Ch 2)
  • Semantic index — A lexical and semantic index of Microsoft Graph data that lets Copilot retrieve conceptually relevant content while respecting each user's permissions. (Ch 2)
  • Token — The unit an LLM reads and generates — roughly ¾ of an English word; the basis of consumption-based cost. (Ch 1)