L5: Copilot Studio — Hybrid Pattern with Microsoft Foundry

📋 Agenda

Estimated reading time: ~30 minutes | Hands-on Copilot Studio Lab

Learning outcomes:

✅ Distinguish clearly between Copilot Studio and Foundry Workflow Builder (updated decision matrix for New Foundry)
✅ Build a conversational bot with Copilot Studio's Topic Designer
✅ Connect Copilot Studio to a Foundry agent using the hybrid pattern
✅ Deploy to Teams and website widget via multi-channel publishing

Prerequisites:

Microsoft 365 business account with Copilot Studio license
Completed L1 — have a Foundry Prompt Agent with a known endpoint
Copilot Studio license (per-tenant or per-conversation)

1. Problem Statement

1.1. A Clarified Landscape — New Foundry Changed the Boundaries

In Foundry Classic, the division was clear:

Foundry Classic → single agent, no visual flow
Copilot Studio → visual conversation flow, multi-channel, Power Platform connectors

New Foundry has shifted this boundary:

New Foundry Workflow Builder → now also has visual orchestration, sequential flows, human gates
Copilot Studio → retains its unique strengths that Foundry Workflow Builder does NOT yet have

The question "when to use which" requires an updated answer.

1.2. What Copilot Studio Still Does Uniquely

Foundry Workflow Builder excels at AI-to-AI orchestration (agents delegating to agents). Copilot Studio excels at conversation flow design (routing users through explicit dialog trees with business logic and Power Platform integration).

2. Architecture: Hybrid Pattern

The most powerful pattern combines both platforms: Copilot Studio as the front door for conversation management, Foundry as the specialized AI engine for complex tasks.

Pattern description: Copilot Studio acts as the receptionist — it handles greeting, navigation, simple form collection, and routing. When a user's request requires genuine AI reasoning (RAG on documents, multi-step analysis), Copilot Studio delegates to a Foundry agent and returns the response to the user.

3. Glossary & Vocabulary

Term	Vietnamese Meaning	Explanation
Topic	Chủ đề hội thoại	A defined conversational path in Copilot Studio — triggered by specific phrases
Trigger Phrases	Câu kích hoạt	Phrases that cause a Topic to activate (e.g., "I want to check my leave balance")
Node	Nút (trong Topic)	A single step in a Topic: Message, Question, Condition, Action, or Agent call
Power Automate	Tự động hóa Power	Microsoft's workflow automation tool — connects to 1,000+ business apps
Widget	Tiện ích nhúng	A chat window that can be embedded in any website via `<iframe>`

4. Lab A: Building a Topic in Copilot Studio

4.1. Access Copilot Studio

🌐 https://copilotstudio.microsoft.com
  → Sign in with Microsoft 365 account
  → Select your environment
  → Click "+ Create" → "New agent"

Creation options:

Option A — AI-Assisted (describe in natural language):
  "I want to build an internal employee support bot that handles
  HR questions about leave and benefits, IT helpdesk requests,
  and searches internal policy documents from SharePoint."
  → AI generates initial Topics based on your description

Option B — Manual:
  Click "Skip to configure"
  → Name: "Internal Employee Bot"
  → Instructions: [system-level prompt for fallback behavior]
  → Language: Vietnamese

4.2. Topic Designer Canvas

Left sidebar → Topics → "+ New topic" → "From blank"

The Topic Designer:

┌──────────────────────────────────────────────────────────────┐
│  Nodes Panel (left)      │  Canvas (center)                  │
│  ─────────────────────   │  ─────────────────────────────    │
│  Trigger Phrases         │  [Trigger Phrases node]           │
│  Message                 │          ↓                        │
│  Question                │  [Message node]                   │
│  Condition               │          ↓                        │
│  Action (Power Automate) │  [Question node]                  │
│  Call Agent (Foundry)    │       ↙     ↘                     │
│  End Conversation        │  [Path A]  [Path B]               │
└──────────────────────────────────────────────────────────────┘

Lab: Leave Balance Topic

1. [Trigger Phrases node]
   Add:
   - "How many leave days do I have left?"
   - "Check remaining leave"
   - "Leave balance"

2. [Message node]
   "I will check your leave balance. Please provide your employee ID."

3. [Question node]
   Question: "What is your employee ID?"
   Save response as: {employee_id}

4. [Action node → Power Automate]
   Flow: "Get-Leave-Balance"
   Input: {employee_id}
   Output: {leave_days_remaining}

5. [Message node]
   "You have {leave_days_remaining} leave days remaining this year."

6. [End Conversation]

5. Lab B: Connecting Copilot Studio to a Foundry Agent

This is the hybrid pattern — the most powerful integration between the two platforms.

5.1. Get Foundry Agent Endpoint

ai.azure.com → Build → Agents → [Your Agent] → Properties
  → Endpoint: https://your-project.foundry.azure.com/agents/agt_xxxx
  → Agent ID: agt_xxxxxxxxxxxx

5.2. Add Foundry Agent as a Connected Agent

Copilot Studio
  → Settings → Generative AI → "Connected agents"
  → "+ Add" → "Microsoft Foundry"
  → Endpoint URL: https://your-project.foundry.azure.com
  → Agent ID: agt_xxxxxxxxxxxx
  → Test connection → Save

5.3. Create a Topic That Delegates to Foundry

Topics → "+ New topic" → "Complex Technical Query"

Trigger Phrases:
  - "I have a technical question"
  - "Search policy documents"
  - "I need detailed analysis"

Flow:
1. [Trigger]
2. [Question] "What do you need help with?"
   → Save to: {user_query}
3. [Call Agent — Foundry]
   Agent: your-foundry-agent
   Input: {user_query}
   Output: {agent_response}
4. [Message] "{agent_response}"
5. [End]

Result: Copilot Studio handles simple topics natively (HR, IT navigation, Power Automate actions). For complex AI reasoning, it seamlessly delegates to the Foundry agent and returns the AI-generated response to the user — all within the same Teams or website chat interface.

6. Multi-Channel Publishing

6.1. Publish to Teams

Copilot Studio → Publish → "Microsoft Teams"
  → Fill: App Name, Description, Icon
  → Click "Publish"
  → IT Admin approves in Teams Admin Center (same process as L4)

Copilot Studio → Channels → "Custom website"
  → Enable: Toggle ON
  → Copy embed code:

<iframe
  src="https://copilotstudio.microsoft.com/environments/.../bots/.../webchat"
  style="width: 400px; height: 600px; border: 0"
  title="Support Bot">
</iframe>

Paste this code into any HTML page — the chat widget appears in the bottom-right corner.

7. Updated Decision Matrix — When to Use What

Scenario	Recommendation
RAG bot answering from uploaded documents	Foundry Prompt Agent
Content pipeline: research → review → write	Foundry Workflow Agent
HR bot with leave form + Dynamics 365 integration	Copilot Studio
Customer support with RAG + SharePoint + website chat	Hybrid
Chatbot embedded on company website	Copilot Studio (widget)
Agent that analyzes uploaded CSV and generates charts	Foundry Prompt Agent (Code Interpreter)

8. Discussion

"New Foundry now has its own visual workflow builder — does Copilot Studio still have a future?"

Different products solving different problems, despite surface-level similarity. Foundry Workflow Builder is optimized for AI-to-AI orchestration: connecting agents that produce structured outputs for each other, with a technical audience in mind (YAML export, VS Code integration, CI/CD). Copilot Studio is optimized for dialog-driven UX authoring: non-technical business users (HR, marketing, operations) designing explicit conversation paths, collecting user input at specific steps, and connecting to Power Platform — a 1,000+ connector ecosystem that has no equivalent in Foundry today. The technical overlap is narrow; the intended user and use case differ substantially. The hybrid pattern described in this article represents the practical answer: use each platform for what it is uniquely good at.

9. Series Summary — Part 5: Low-Code Path

Next steps for Low-Code learners:

Explore Power Automate to build automation workflows connecting your Copilot Studio agent to business systems
Try Microsoft Graph connectors to let Foundry agents read from email, calendar, and SharePoint automatically
Investigate Foundry Hosted Agents (L0 concept) for cases where you need custom Python code but want Foundry's managed infrastructure

Made by Anh Tu - Share to be shared

📋 Agenda​

Learning outcomes:​

Prerequisites:​

1. Problem Statement​

1.1. A Clarified Landscape — New Foundry Changed the Boundaries​

1.2. What Copilot Studio Still Does Uniquely​

2. Architecture: Hybrid Pattern​

3. Glossary & Vocabulary​

4. Lab A: Building a Topic in Copilot Studio​

4.1. Access Copilot Studio​

4.2. Topic Designer Canvas​

5. Lab B: Connecting Copilot Studio to a Foundry Agent​

5.1. Get Foundry Agent Endpoint​

5.2. Add Foundry Agent as a Connected Agent​

5.3. Create a Topic That Delegates to Foundry​

6. Multi-Channel Publishing​

6.1. Publish to Teams​

6.2. Publish Website Widget​

7. Updated Decision Matrix — When to Use What​

8. Discussion​

9. Series Summary — Part 5: Low-Code Path​