Introduction.

Welcome to the Agent Development Workshop! This hands-on, code-first journey teaches you how to build, orchestrate, evaluate, and deploy production-grade AI agents using Google’s Agent Development Kit (ADK).

Before You Begin

Ensure your workspace is synchronized with the starting state for this module:

make catchup module=00

[!NOTE] This command safely resets your workspace/ to the known good starting point for Module 00.

---

1. Welcome to the Workshop

Developing software with LLMs is transitioning from simple prompt engineering to engineering autonomous systems. When we build applications where the AI makes decisions, calls tools, and coordinates with other AI systems, we need reliable framework patterns, local debugging tools, and robust evaluation loops.

This workshop is designed to teach you those exact patterns. You will act as a developer building a sophisticated backend and pairing it with a web interface to deliver a polished user experience.

2. What are Agents?

An AI Agent is an autonomous entity powered by a Large Language Model (LLM) that can:

1. Reason and Plan: Break down complex goals into discrete sub-tasks.
2. Use Tools: Query external APIs, search databases, write code, or perform calculations.
3. Maintain State: Keep track of conversation history, user preferences, and intermediate results.
4. Collaborate: Transfer tasks or delegate sub-problems to other specialized agents.

Unlike traditional software that follows rigid conditional branches, agents leverage the reasoning capabilities of LLMs to dynamically decide the best path to accomplish a task.

3. What is the Agent Development Kit (ADK)?

The Agent Development Kit (ADK) is Google’s open-source, code-first framework for building, testing, evaluating, and deploying autonomous agents. It provides a developer-friendly Python SDK and CLI that abstracts the complexities of state preservation, tool-calling loop execution, and multi-agent orchestration.

Key primitives in ADK 2.0 include:

• Agent (LlmAgent): The core intelligent actor defined by an instruction, a model, and a set of tools.
• Tool: A regular Python function decorated or bound to an agent, giving it real-world capabilities.
• Workflow: A graph-based execution pipeline of nodes and edges for deterministic multi-agent coordination.
• Runner: The execution engine that runs agents and workflows, managing event emission and lifecycle callbacks.
• Session & State: Stateful storage for maintaining conversational history and short-term agent memory.

4. What We’ll Build: The AI Newsroom

Throughout this workshop, you will construct a fully-featured AI Newsroom application. It allows users to input a broad research topic, which then triggers a pipeline of specialized agents:

1. Planner Agent: Parses the user request and plans specific web search queries.
2. Researcher Agents: Conduct parallel web searches using Google Search Grounding to find reliable information.
3. Librarian Agent: Searches an archive of local documents using Vertex AI Vector Search.
4. Editor-in-Chief Agent: Merges the web research and archive results into a unified, formatted newspaper page.

This backend connects to a beautiful web application interface. Here is what the application looks like:

Newsroom Home

Students and users enter their search prompt, choose a research model, and customize settings.

Real-Time Researching

The UI streams the step-by-step thinking, search queries, and results from the agents in real-time.

Polished Newsletter Output

The final result is rendered in a clean grid layout featuring headlines, articles, and grounded citations.

---

Getting unstuck

If you fall behind or need a clean start for any module, you can instantly catch up to the beginning of the module by running the following command in your terminal:

make catchup module=01

(Replace 01 with the module number you wish to start).

---

Resources

• ADK Docs (adk.dev): The official website and documentation for the Google Agent Development Kit.
• ADK Python GitHub: The open-source repository for the ADK Python SDK.
• Google GenAI SDK: The underlying client library used to interact with Gemini models.