AI Agents for Beginners - 2. Explore Agentic Frameworks

AI Agents for Beginners - 2. Explore Agentic Frameworks

Source: microsoft/ai-agents-for-beginners - 02-explore-agentic-frameworks

What are AI Agent Frameworks?

How to Prototype and Iterate Quickly

Modular Components

modular_agent.py
python

modular_agent.py Flow
Mermaid
flowchart TD
    User["User\n'Go to New York on Jan 1, 2025'"]
    --> Provider["AzureAIProjectAgentProvider\ncreate_agent(tools=[book_flight])"]
    Provider --> Agent["travel_agent"]
    Agent --> LLM["LLM\nAnalyze request → decide to call book_flight"]
    LLM --> Tool["book_flight(date, location)"]
    Tool --> Agent2["Agent\nReceive result → generate response"]
    Agent2 --> Response["'Travel was booked to New York on Jan 1'"]

Collaborative Tools

multi_agent.py
python

multi_agent.py Flow
Mermaid
flowchart TD
    Task["Task: Retrieve sales data for Q4"]
    --> AgentR["dataretrieval agent\ntool: retrieve_tool"]
    AgentR --> R1["retrieval_result"]
    R1 --> AgentA["dataanalysis agent\ntool: analyze_tool"]
    AgentA --> R2["analysis_result\n(final insights)"]

Real-Time Learning

Microsoft Agent Framework vs Azure AI Agent Service

Microsoft Agent Framework (MAF)

maf_agent.py
python

maf_agent.py Flow
Mermaid
flowchart LR
    Setup["create_agent(name, instructions)"]
    --> Agent["Agent instance\n(my_agent)"]
    Agent --> Run["agent.run('Hello, World!')"]
    Run --> Response["Response text"]

maf_tool.py
python

maf_tool.py Flow
Mermaid
flowchart TD
    Func["Define get_weather(location)"]
    --> Register["create_agent(tools=[get_weather])"]
    Register --> Agent["weather_agent"]
    User["User message"] --> Agent
    Agent --> LLM["LLM\nDecide whether to call tool"]
    LLM -->|"Call automatically if needed"| Func2["get_weather(location)"]
    Func2 --> Agent
    Agent --> Response["Response"]

maf_multi_agent.py
python

maf_multi_agent.py Flow
Mermaid
flowchart TD
    Input["'Plan a trip to Paris'"]
    --> Planner["planner agent\nBreak down complex tasks into steps"]
    Planner --> Plan["plan (step-by-step text)"]
    Plan --> Executor["executor agent\nExecute steps using tools"]
    Executor --> Result["Final result"]

Azure AI Agent Service

azure_ai_agent_service.py
python

azure_ai_agent_service.py Flow
Mermaid
zenuml
    title azure_ai_agent_service.py Flow
    Client->AgentService: create_agent(model, name, tools)
    Client->AgentService: create_thread()
    Client->AgentService: create_message(thread_id, user_input)
    Client->AgentService: create_and_process_run(thread_id, agent_id)
    AgentService->LLM: process request with tools
    LLM->AgentService: tool calls and response
    AgentService->Client: list_messages() returns response

• Maintains conversation state (context) across multiple turns
• Responses can take various forms, including text, images, and files
• Operates seamlessly integrated with MAF (AzureAIProjectAgentProvider)

Which One to Use?

MAF vs Azure AI Agent Service
Mermaid
flowchart TD
    Start["What do you need?"] --> Q1{Quick start\nwith simple API?}
    Q1 -->|Yes| MAF["Microsoft Agent Framework\nAzureAIProjectAgentProvider\nRapid prototyping"]
    Q1 -->|No| Q2{Enterprise deployment\nor Azure integrations?}
    Q2 -->|Yes| AAAS["Azure AI Agent Service\nAzure Search / Bing / Functions\nEnterprise scale"]
    Q2 -->|No| Q3{Multi-model support?\nLlama / Mistral / Cohere}
    Q3 -->|Yes| AAAS
    Q3 -->|No| MAF
    MAF -->|"Prototype done"| Both["Deploy via\nAzure AI Agent Service"]
    AAAS --> Both

Summary

Lesson 02 Summary
Mermaid
flowchart LR
    Root["Agentic Frameworks"]
    Root --> Why["Why Frameworks?"]
    Root --> Prototype["Rapid Prototyping"]
    Root --> Compare["MAF vs AAAS"]
    Why --> W1["Agent Collaboration"]
    Why --> W2["Task Automation"]
    Why --> W3["Contextual Adaptation"]
    Prototype --> P1["Modular Components"]
    Prototype --> P2["Collaborative Tools"]
    Prototype --> P3["Real-Time Learning"]
    Compare --> M1["MAF\nSDK / Rapid development"]
    Compare --> M2["AAAS\nPlatform / Enterprise"]
    Compare --> M3["Recommended to use together"]

• AI Agent Frameworks go beyond traditional AI to enable autonomous multi-agent systems
• Iterate and improve quickly with Modular Components, Collaborative Tools, and Real-Time Learning
• MAF is optimized for rapid development, while Azure AI Agent Service is optimized for enterprise deployment
• Using both together is the recommended strategy