Skip to content

02 Develop an AI agent with Azure AI Foundry Agent Service

1. 🧩 Problem

How can you build a scalable, secure, and customizable AI agent in Azure to automate complex tasks (e.g., customer service, report generation, workflow automation) without managing infrastructure or writing excessive amounts of code?

2. 🚀 Solution with Azure

Use Azure AI Foundry Agent Service — a fully managed service that enables developers to:

  • Build, deploy, and scale high-quality AI agents
  • Leverage generative AI models, integrated tools, and custom functions
  • Avoid managing compute or storage resources

3. 🧱 Required Components

Component Description
✅ Azure AI Hub Central workspace for managing AI assets
✅ Azure AI Project Scoped environment for building and managing agents
✅ Deployed AI Model e.g., Azure OpenAI, Llama 3, Mistral
✅ Azure AI Services Underlying cognitive and ML services
➕ Optional:
• Azure Key Vault Secure secrets and credentials
• Azure AI Search Enable semantic search for knowledge grounding
• Azure Storage External file access and management

Agent Tools:

  • Knowledge Tools: Bing Search, File Search, Azure AI Search, Microsoft Fabric
  • Action Tools: Code Interpreter, Azure Function, Custom Function, OpenAPI Spec

4. 🏗️ Architecture / Development

🔁 Development Flow

  1. Connect to Azure AI Foundry Project
  2. Use project endpoint

  3. Authenticate via Microsoft Entra ID

  4. Create or Reference an Agent

  5. Specify model deployment
  6. Define behavior using Instructions

  7. Assign tools and resources

  8. Create a Thread (stateful conversation container)

  9. Send Messages and invoke the agent
  10. Check Status and retrieve responses and generated artifacts
  11. Repeat the conversation loop
  12. Delete the agent and thread to clean up resources

🛠️ Development Options

  • Azure AI Foundry Portal (low-code)
  • SDKs / REST APIs (e.g., Python)
  • Bicep templates (for provisioning resources)

5. ✅ Best Practices / Considerations

  • 🔐 Use threads to securely manage conversation state
  • 🧰 Assign tools that match agent responsibilities
  • 🔑 Prefer keyless authentication for secure access
  • ☁️ Use your own Azure Blob Storage if full control is required
  • 🔄 Use Copilot Studio or Semantic Kernel for:
  • Multi-agent orchestration
  • Microsoft 365 integration

6. 📝 Sample Exam Questions

Q1. What are the key benefits of using Azure AI Foundry Agent Service over building agents manually via APIs?
A1. Simplifies development, requires less code, manages infrastructure, and includes built-in tools like Code Interpreter and Azure Functions.

Q2. Which component manages the state and context of an AI agent's conversation in Foundry?
A2. Thread

Q3. When would you choose Copilot Studio or Semantic Kernel over Foundry Agent Service?
A3. When integrating with Microsoft 365 or orchestrating multiple agents.

Q4. What’s the difference between a knowledge tool and an action tool?
A4. Knowledge tools provide grounding data (e.g., Bing Search); action tools perform tasks (e.g., Azure Function).

Q5. Which models are compatible with Foundry Agent Service?
A5. Azure OpenAI, Llama 3, Mistral, Cohere