Certification Workshop – Intelligent Automation: Building Collaborative AI Agents

• Introduction: The "why" behind agentic AI.
• Core Concepts: What is an AI Agent? The magic formula: LLM (Brain) + Tools (Hands) + Task (Goal).
• The Landscape: Quick overview of popular frameworks (e.g., CrewAI, LangGraph).
• Introducing the MVP: We will build a Logistics & Supply Chain System to automate an order fulfillment process. We'll introduce our three specialist agents:
  
  • Supplier Agent: Manages raw material inventory.
  • Manufacturer Agent: Manages production orders.
  • Logistics Agent: Manages shipment and delivery.

• Goal: Create and test each agent in isolation to understand their core capabilities.
• MVP v1 Scope:
  
  • Setup: Participants will get a pre-configured code template.
  • Supplier Agent:
    
    • Task: Check inventory for a specific part.
    • Tool: Connect a pre-built Python function to read from an inventory.csv file.
  • Manufacturer Agent:
    
    • Task: Check the status of a production order.
    • Tool: Connect a tool to query a simple SQLite orders.db database.
  • Logistics Agent:
    
    • Task: Get a shipping quote for a destination.
    • Tool: Connect a tool that calls a mock API endpoint to get a delivery price.
  • Outcome: Participants can run each agent individually with a specific task (e.g., "Is part #A123 in stock?") and see it use its tool to get a correct answer.

• Goal: Introduce the concept of shared context and set up the central "MCP Server".
• Limitations of MVP v1.
• Introducing MCP: We will define Model Context Protocol (MCP) as our strategy for a shared "whiteboard" or "workspace." This is a central server where agents can read/write data, share files, and see each other's status.
• Hands-on (MVP v2 Setup):
  
  • Participants will start a pre-built, simple FastAPI server locally. This is our MCP Server.
  • They will upgrade their agents' tools. The Supplier Agent will now read the inventory.csv from a URL endpoint on the MCP server, not a local file.
  • Outcome: The agents are still triggered manually, but they now fetch their resources from a centralized location, preparing them for true collaboration.

Take a breather, refresh and connect with other participants.

• Goal: Connect the agents into a single, automated workflow where they trigger each other to complete a complex task.
• Defining Agent-to-Agent (A2A) communication. How an agent can delegate a task to another agent in the team. We will implement this as a new "delegation" tool.
• Hands-on (MVP v3 Build):
  
  • The main task is now: "Process new order #555". This task is given only to the Manufacturer Agent.
  • The Manufacturer Agent's logic is upgraded. When it checks the database and sees it needs parts, it won't just report it. It will use a new delegate_task tool to activate the Supplier Agent with the instruction: "Procure parts for order #555."
  • The Supplier Agent is upgraded. After procuring parts, it will delegate a task to the Logistics Agent: "Schedule shipment for the parts for order #555."
• Participants will run a single command to kick off the entire workflow and watch the agents communicate and collaborate in the terminal, providing status updates as they complete their parts of the job.
• Outcome: A fully functional, collaborative multi-agent system that autonomously handles a process from start to finish.

• Review the complete agentic workflow we just built.
• Best Practices: Discussing the next steps:
  
  • Error handling and validation.
  • Adding a Human-in-the-Loop for approval steps.
  • Security considerations for agent tools.
  • Scaling the MCP server.
• Advanced use cases (e.g., dynamic agent creation, financial analysis teams) to inspire further building.

• Get access to the final code, slides, and a list of further resources
• Connect with peers & mentors and receive your certificate of participation.