Once Upon Agentic AI: A Developer's Epic Journey into Building End-to-end Agents using Open Source Strands

This workshop provides a structured, hands-on journey through the end-to-end process of building and deploying multi-agent systems using the Strands Agents - an open-source agent framework.

Framed around building an AI Game Master for a fantasy role-playing game in Python, participants begin by constructing a foundational conversational agent with system prompts and debug logs. They then progressively unlock advanced capabilities, including integrating pre-built SDK tools (such as HTTP requests), forging custom Python @tool functions, exposing and consuming tools via a Model Context Protocol (MCP) server, and orchestrating specialized remote sub-agents using the Agent-to-Agent (A2A) protocol. The workshop also addresses critical production considerations such as implementing Human-in-the-Loop safety gates and enforcing structured outputs. We will also learn about building and agent using harness architecture and deploying the entire multi-agent architecture with communication, managed knowledge bases, and persistent memory.

By the end of the session, participants will have built and deployed a fully functional, production-ready multi-agent application and gained practical experience with core patterns and best practices for agentic AI development.

Workshop Infrastructure

Attendees will be provided with the cloud infrastructure and access to cloud models to run the workshop.

Duration

2-3 hours

Workshop Repository

https://github.com/aws-samples/sample-once-upon-agentic-ai

Attendees will learn:

1. How to initialize foundational AI agents with the Strands Agents SDK, craft system prompts, and inspect agent reasoning via debug logs.

2. How to equip agents with pre-built tools (web fetch, shell, file editing) gated by HumanInTheLoop safety controls, as well as custom Python tools using the @tool decorator.

3. How to decouple and serve tools across boundaries by building a FastMCP server and configuring agents to consume tools over the Model Context Protocol (MCP).

4. How to design and orchestrate multi-agent systems using the Agent-to-Agent (A2A) protocol, connecting a Game Master orchestrator with remote Rules and Character sub-agents.

5. How to enforce structured outputs and streamline agent APIs using the Strands harness and an interactive web interface.

6. How to persist state and ground agent responses.

7. How to deploy multi-agent workflows and MCP/A2A servers to production.