AI Agents for Beginners - 8. Multi-Agent Design Pattern

AI Agents for Beginners - 8. Multi-Agent Design Pattern

Source: microsoft/ai-agents-for-beginners - 08-multi-agent

Scenarios Where Multi-Agents Are Applicable

Advantages of Using Multi-Agents Over a Singular Agent

Single Agent vs Multi-Agent
Mermaid
flowchart LR
    subgraph Single["Single Agent (Mom & Pop)"]
        SA["SingleAgent\nHandles flights + hotels\n+ car rentals all-in-one"]
    end
    subgraph Multi["Multi-Agent (Franchise)"]
        FA["FlightAgent"]
        HA["HotelAgent"]
        CA["CarAgent"]
    end
    User1["User"] --> SA
    User2["User"] --> FA & HA & CA

Building Blocks of Implementing the Multi-Agent Design Pattern

Multi-Agent Building Blocks
Mermaid
flowchart TD
    MAD["Multi-Agent\nDesign Pattern"]
    MAD --> Comm["Agent Communication\nInformation sharing method"]
    MAD --> Coord["Coordination\nConstraint alignment"]
    MAD --> Arch["Agent Architecture\nDecision-making & learning"]
    MAD --> Vis["Visibility\nActivity tracking"]
    MAD --> Pat["Patterns\nGroup/Hand-off/Filter"]
    MAD --> HITL["Human-in-the-Loop\nDefine intervention points"]

Multi-Agent Patterns

Group Chat

• Can be implemented as a Centralized architecture via a central server, or a Decentralized architecture where agents exchange messages directly
• Each agent sends, receives, and responds to messages in the group chat

Group Chat Pattern
Mermaid
flowchart LR
    GroupChat["Group Chat\n(Central Hub)"]
    A1["Agent 1\nTravelPlanner"] <--> GroupChat
    A2["Agent 2\nConcierge"] <--> GroupChat
    A3["Agent 3\nBudgetReviewer"] <--> GroupChat
    User["User"] --> GroupChat
    GroupChat --> User

Hand-off

Hand-off Pattern
Mermaid
zenuml
    title Hand-off Pattern
    User->AgentA: initial request
    AgentA->AgentA: process own scope
    AgentA->AgentB: hand off with context
    AgentB->AgentB: process own scope
    AgentB->AgentC: hand off with context
    AgentC->User: final response

Collaborative Filtering

Collaborative Filtering Pattern
Mermaid
flowchart TD
    User["User\n'Which stock should I buy?'"]
    --> Coordinator["Coordinator Agent\nDispatch requests & aggregate results"]

    Coordinator --> IE["Industry Expert Agent\nIndustry trends & sector analysis"]
    Coordinator --> TA["Technical Analysis Agent\nCharts, patterns & entry point analysis"]
    Coordinator --> FA["Fundamental Analysis Agent\nFinancial statements & company valuation"]

    IE --> Agg["Aggregator\nSynthesize multi-angle insights"]
    TA --> Agg
    FA --> Agg

    Agg --> Rec["Comprehensive Recommendation\nHolistic investment recommendation"]
    Rec --> User

Creating Specialized Agents

Setup

setup.py
python

specialized_agents.py
python

Specialized Agents
Mermaid
flowchart LR
    P["TravelPlanner\nDraft itinerary\n(attractions, logistics)"]
    C["TravelConcierge\nReview & enhance plan\n(tips, restaurants, issues)"]
    P -->|"Pass context"| C

Sequential Workflow with WorkflowBuilder

sequential_workflow.py
python

Sequential Workflow: 2-Agent Pipeline
Mermaid
flowchart LR
    U["User\n'Plan 5-day Paris, $3000'"]
    --> W["WorkflowBuilder\nstart_executor=planner_agent"]
    W --> P["TravelPlanner\nDraft itinerary\n(attractions, logistics)"]
    P -->|"AgentResponseUpdate"| C["TravelConcierge\nReview & enhance plan\n(tips, restaurants, issues)"]
    C -->|"AgentResponseUpdate"| S["Stream to User"]

Extending the Workflow

extended_workflow.py
python

Extended 3-Agent Pipeline
Mermaid
flowchart LR
    U["User\n'5-day Paris, $3000'"]
    --> P["TravelPlanner\nDraft itinerary"]
    --> C["TravelConcierge\nReview & enhance"]
    --> B["BudgetReviewer\nBudget validation & savings suggestions"]
    --> R["Final\nItinerary"]

Visibility into Multi-Agent Interactions

Scenario: Refund Process

Refund Process Multi-Agent System
Mermaid
%%{init: {'look': 'handDrawn', 'themeVariables': {'clusterBkg': '#e8efff28', 'clusterBorder': '#aabbcc'}, 'flowchart': {'subGraphTitleMargin': {'top': 14, 'bottom': 8}}}}%%
flowchart TD
    subgraph Core["Core Refund Flow (Sequential)"]
        Customer["Customer Agent\nRefund request"]
        Resolution["Resolution Agent\nIssue analysis & coordination"]
        Seller["Seller Agent\nRefund approval"]
        Payment["Payment Agent\nPayment refund"]
    end

    subgraph Checks["Parallel Validation — Resolution Stage"]
        direction LR
        Compliance["Compliance Agent\nCompliance verification"]
        Security["Security Agent\nSecurity verification"]
        Knowledge["Knowledge Agent\nPolicy & rules reference"]
    end

    subgraph Fulfillment["Parallel Fulfillment — Seller Stage"]
        Shipping["Shipping Agent\nReturn shipment processing"]
    end

    subgraph PostProcess["Post-Processing (Parallel)"]
        direction LR
        Feedback["Feedback Agent\nCollect customer feedback"]
        Analytics["Analytics Agent\nData analysis"]
        Audit["Audit Agent\nProcess audit"]
        Reporting["Reporting Agent\nGenerate reports"]
        Quality["Quality Agent\nQuality verification"]
    end

    Escalation["Escalation Agent\nEscalate to higher support"]
    Notification["Notification Agent\nStep-by-step customer notifications"]

    Customer --> Resolution
    Resolution -->|"Complex issue"| Escalation
    Resolution --> Compliance & Security & Knowledge
    Compliance & Security & Knowledge --> Seller
    Seller --> Payment & Shipping
    Customer & Seller & Payment --> Notification
    Payment --> Feedback & Analytics & Audit & Reporting & Quality

Summary

Lesson 08 Summary
Mermaid
flowchart LR
    Root["Multi-Agent\nDesign Pattern"]
    Root --> Why["Why Multi-Agent?"]
    Root --> Blocks["Building Blocks"]
    Root --> Patterns["Patterns"]
    Root --> Code["WorkflowBuilder"]
    Root --> Vis["Visibility"]
    Why --> W1["Specialization"]
    Why --> W2["Scalability"]
    Why --> W3["Fault Tolerance"]
    Blocks --> B1["Communication"]
    Blocks --> B2["Coordination"]
    Blocks --> B3["Architecture"]
    Patterns --> P1["Group Chat"]
    Patterns --> P2["Hand-off"]
    Patterns --> P3["Collaborative\nFiltering"]
    Code --> C1["create_agent x N"]
    Code --> C2["WorkflowBuilder\n+ add_edge"]
    Code --> C3["stream output"]
    Vis --> V1["Logging"]
    Vis --> V2["Visualization"]
    Vis --> V3["Metrics"]

• Multi-Agent Systems overcome the limitations of single agents through Specialization, Scalability, and Fault Tolerance.
• Connect agents into sequential pipelines using WorkflowBuilder's add_edge, and scale by adding new agents without modifying existing code.
• Choose agent communication and coordination methods using three core patterns: Group Chat, Hand-off, and Collaborative Filtering.
• As the number of agents grows, securing interaction visibility through Logging, Visualization, and Performance Metrics becomes crucial.