5.6 Microsoft Copilot and AI-Powered Applications
Key Takeaways
- Microsoft Copilot is a GPT-powered AI assistant integrated into Microsoft 365, Windows, Edge, GitHub, Dynamics 365, and Power Platform.
- In Microsoft 365, Copilot drafts emails in Outlook, writes in Word, analyzes data in Excel, builds decks in PowerPoint, and summarizes meetings in Teams.
- Copilot uses a RAG pattern over your data via Microsoft Graph, keeping enterprise data inside the Microsoft 365 boundary.
- Azure AI Foundry (formerly Azure AI Studio) is the unified portal for building, evaluating, and deploying custom AI apps, with hubs and projects.
- Custom AI solutions combine Azure services — e.g., Azure OpenAI + Azure AI Search for grounded chatbots, or Vision + Custom Vision for visual inspection.
Quick Answer: Microsoft Copilot is an AI assistant integrated into Microsoft 365, Windows, Edge, and other products. It uses GPT models with RAG over your data to help draft emails, build presentations, analyze data, and summarize meetings. Azure AI Foundry is the portal for building custom AI applications. The AI-900 tests awareness of these capabilities.
Microsoft Copilot
Microsoft Copilot is an AI-powered assistant embedded across Microsoft's product ecosystem. It uses large language models (the GPT family) combined with RAG over your data to provide contextual assistance grounded in your content.
Copilot in Microsoft 365
| Application | What Copilot Does | Example |
|---|---|---|
| Outlook | Draft emails, summarize threads, suggest replies | "Draft a polite reply declining this meeting" |
| Word | Write, rewrite, summarize, and format documents | "Summarize this 20-page report into key findings" |
| Excel | Analyze data, build formulas, generate charts, spot trends | "What are the top 5 products by revenue?" |
| PowerPoint | Create presentations from prompts or documents | "Create a deck from this Word document" |
| Teams | Summarize meetings, capture action items, answer questions | "What decisions were made in today's meeting?" |
| OneNote | Summarize notes, generate to-do lists, draft content | "Create a to-do list from my meeting notes" |
How Copilot Works (The Architecture)
- User prompt — you type a request in natural language.
- Grounding — Copilot searches your organization's data (emails, documents, chats, calendars) for relevant context using Microsoft Graph.
- Augmented prompt — your request + retrieved context + responsible-AI instructions are combined.
- LLM processing — a GPT model generates a response grounded in your data.
- Post-processing — content filters check the response and citations are added.
- Response — the grounded, filtered answer appears in the application.
This is a RAG pattern: retrieve (Microsoft Graph) -> augment (build the prompt) -> generate (GPT). Knowing this clarifies why Copilot can answer about your meetings and files while a public chatbot cannot — and why your enterprise data stays inside the Microsoft 365 boundary.
Copilot in Other Products
| Product | What Copilot Does |
|---|---|
| Windows | System tasks, file search, settings help, content creation |
| Edge | Summarize web pages, generate content, answer questions about a page |
| GitHub | Code completion, code review, documentation, bug detection |
| Dynamics 365 | CRM insights, sales recommendations, customer summaries |
| Power Platform | Build apps and automations from natural-language descriptions |
Azure AI Foundry
Azure AI Foundry (formerly Azure AI Studio) is the unified portal for building, evaluating, and deploying custom AI applications:
| Feature | Description |
|---|---|
| Model Catalog | Browse and deploy models from OpenAI, Meta, Mistral, Cohere, and more |
| Prompt Flow | Visual builder for AI application workflows |
| Evaluation | Tools to assess quality, groundedness, and safety |
| Content Safety | Configure content filters and guardrails |
| Deployments | Manage model deployments with version control |
| Data and Indexes | Connect data sources for RAG and fine-tuning |
| Agent Builder | Build agents that call tools and take actions |
Within Foundry, a hub is the top-level container for shared resources (compute, storage, connections), and a project is a workspace within a hub for a specific AI application.
Building AI Applications on Azure
Organizations combine Azure AI services to build custom solutions. Common patterns:
| Pattern | Description | Azure Services |
|---|---|---|
| Smart chatbot | Natural-language customer service bot | Azure OpenAI + Azure AI Search + Bot Service |
| Document processing | Automated data extraction from forms | Document Intelligence + Azure AI Search |
| Content moderation | Review user-generated content | Azure AI Content Safety |
| Search and discovery | Intelligent search over org knowledge | Azure AI Search + Azure OpenAI |
| Real-time translation | Multilingual communication platform | Azure AI Speech + Azure AI Translator |
| Visual inspection | Automated quality control from images | Azure AI Vision + Custom Vision |
The AI Application Stack (conceptual layers)
End-user application (web/mobile/chatbot/Copilot) -> orchestration layer (Prompt Flow, Semantic Kernel) -> AI models (Azure OpenAI, Azure AI services) -> data layer (Azure AI Search, SQL, Cosmos DB) -> infrastructure (subscriptions, RBAC, networking).
Copilot vs. Building Your Own AI Application
A common point of confusion is the difference between using Copilot and building a custom AI application:
| Microsoft Copilot | Custom application (Azure AI Foundry / Azure OpenAI) | |
|---|---|---|
| Who builds it | Microsoft (it is a finished product) | Your developers |
| How you get it | Buy/enable a Copilot license | Provision and code Azure AI services |
| Data grounding | Automatic via Microsoft Graph | You build the RAG pipeline (Azure AI Search) |
| Customization | Limited to product settings and plugins/agents | Full control over models, prompts, data, and flow |
| Best when | You want productivity in M365 apps fast | You need a tailored experience for your own scenario |
In short, Copilot is a ready-made assistant embedded in products you already use, while Azure AI Foundry is where you build your own assistant from the underlying services. Microsoft also offers Copilot Studio, a low-code tool for building custom copilots and agents that extend Copilot or stand alone — useful when you want more customization than Copilot's defaults but less coding than a full Foundry application.
Knowing where each option sits on the build-vs-buy spectrum answers many "which approach should this organization take?" questions: pick Copilot for fast in-app productivity, Copilot Studio for low-code custom copilots, and Azure AI Foundry plus Azure OpenAI for fully bespoke applications.
On the Exam: Know that Copilot uses RAG over organizational data via Microsoft Graph to give grounded responses, and that Azure AI Foundry is the portal for building custom AI applications. You do not need to build applications — just recognize the concepts and which services solve which problem.
How does Microsoft Copilot in Microsoft 365 access your organizational data to provide relevant responses?
What is Azure AI Foundry?
Which Copilot capability allows it to summarize a meeting you missed and list the action items?
A company wants a customer-service chatbot that answers naturally and is grounded in its own product manuals. Which combination of Azure services fits this pattern best?
Match each Microsoft Copilot application to what it can do:
Match each item on the left with the correct item on the right