Skip to content

Create Custom Agents

This section of the Microsoft AI-102: Designing and Implementing a Microsoft Azure AI Solution exam covers creating and managing custom agents using Azure AI Foundry. Below are study notes for each sub-topic, with links to Microsoft documentation, exam tips, and key facts.

Understand the Role and Use Cases of an Agent

📖 Docs: Overview of Azure AI Agents

Overview

  • Agents are AI-powered entities designed to perform tasks, automate workflows, and interact with users or systems
  • Use cases include:
    • Conversational assistants
    • Enterprise copilots
    • Automation of repetitive processes
    • Knowledge-grounded task execution

Key Points

  • Agents can call external APIs, databases, and integrate with business logic
  • Unlike static LLMs, agents can maintain context across workflows
  • Support single-agent or multi-agent setups

Exam Tip

Expect scenario questions: Which solution requires an agent vs. a standard LLM call?

Configure the Necessary Resources to Build an Agent

📖 Docs: Provision resources for AI agents

Overview

  • Requires setting up an Azure AI Foundry hub/project
  • Linked resources:
    • Azure AI Search (for grounding)
    • Azure Storage (for data persistence)
    • Azure OpenAI (for language models)
  • Networking and RBAC must be configured to ensure security

Key Points

  • Quotas apply to Azure OpenAI (tokens/minute, requests/minute).
  • Agents often require both vector search and LLM endpoints.

Limits

Regional availability and subscription quotas can restrict deployments.

Create an Agent with the Azure AI Foundry Agent Service

📖 Docs: Create an agent

Overview

  • Use the Azure AI Foundry Agent Service to build, configure, and deploy agents
  • Agents can be configured with:
    • Tools (APIs, functions)
    • Memory (short-term, long-term)
    • Prompt templates
    • Grounding sources

Key Points

  • Agents can be hosted and exposed as APIs
  • Agents inherit hub/project security
  • Managed service handles scaling and monitoring

Use Case

Building an enterprise helpdesk copilot that integrates with ticketing systems.

Implement Complex Agents with Semantic Kernel and Autogen

📖 Docs: Semantic Kernel | Autogen

Overview

  • Semantic Kernel (SK): SDK for orchestrating AI skills, plugins, and agents
  • Autogen: Framework for multi-agent conversations, collaboration, and workflows
  • Together, they enable advanced orchestration and complex reasoning

Key Points

  • SK supports chaining skills, context memory, and plugin integration
  • Autogen supports autonomous collaboration between multiple agents
  • Suitable for scenarios like research copilots, coding assistants, or negotiation bots

Exam Tip

Know the difference: - Agent Service = managed, low-code. - Semantic Kernel / Autogen = developer SDKs for custom logic.

Implement Complex Workflows Including Orchestration for Multi-Agent Solutions

📖 Docs: Multi-agent orchestration

Overview

  • Multi-agent = collaboration between multiple specialized agents
  • Supports:
    • Multiple users
    • Multiple tools
    • Autonomous reasoning and task delegation

Key Points

  • Agents can pass tasks between each other
  • Requires orchestration (Semantic Kernel, Autogen, or Foundry workflows)
  • Useful for enterprise-scale copilots or automation across departments

Use Case

  • Travel planning assistant: one agent handles booking, another manages itineraries, another answers customer queries.

Test, Optimize and Deploy an Agent

📖 Docs: Test and deploy agents

Overview

  • Testing ensures groundedness, reliability, and safety
  • Optimize prompts, tools, and grounding sources before production deployment
  • Deployment exposes agents as endpoints for apps/services

Key Points

  • Use evaluation datasets for reliability testing
  • Agents can be deployed via CI/CD
  • Monitoring includes usage, latency, and safety incidents

Best Practices

  • Always include guardrails (content filters, prompt templates)
  • Monitor with Azure Monitor and logging integrations

Quick‑fire revision sheet

  • 📌 Agents are context-aware, unlike static prompts
  • 📌 Requires Azure AI Foundry hub/project + linked resources
  • 📌 Know when to use Agent Service vs. Semantic Kernel/Autogen
  • 📌 Multi-agent = orchestration and autonomous workflows
  • 📌 Testing & deployment require monitoring, evaluation, and guardrails