Google Agent Development Kit (ADK) Workshop

Build AI agents with Google's Agent Development Kit! In this workshop, you'll progressively build a Research Agent, starting from a simple single agent and evolving to a multi-agent hierarchical system.

Workshop Overview

Exercise	Description	Skills Learned
Exercise 1	Simple Single Agent	ADK basics, Agent class, deployment
Exercise 2	Agent with Tools	Custom function tools, web scraping
Exercise 3	Multi-Agent Research Team	AgentTool pattern, orchestration, sub-agents
Exercise 4	Agent Evaluation	Eval framework, rubric-based metrics, prompt engineering

Prerequisites

• Python 3.10+
• Google API Key (from AI Studio)
• Git

Quick Start

1. Clone and Setup

# Clone the repository
git clone <repository-url>
cd adk-workshop-q126

# Create your workshop branch (use your name)
git checkout -b workshop/<your-name>

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Copy environment template and add your API key
cp .env.example .env

Edit .env and add your GOOGLE_API_KEY.

2. Run Locally

adk web

Open http://localhost:8080 to interact with your agent.

3. Deploy

git add .
git commit -m "My research agent"
git push -u origin workshop/<your-name>

Your agent will automatically deploy to Cloud Run.

Your Deployment URL

After pushing, your agent will be available at:

https://<your-name>-adk-workshop-773815123342.us-central1.run.app

For example, branch workshop/sam-gallagher deploys to:

https://sam-gallagher-adk-workshop-773815123342.us-central1.run.app

Exercises

Work through each exercise in order. Instructions are provided during the workshop.

Exercise 1: Simple Single Agent

Create a basic research agent in research_agent/agent.py.

Goal: Define an Agent with a name, model, description, and instruction.

Exercise 2: Agent with Tools

Extend your agent with a custom fetch_webpage tool for web research.

Goal: Build a custom function tool and attach it to an agent.

Exercise 3: Multi-Agent Research Team

Build a research team using the AgentTool pattern where an orchestrator coordinates specialized sub-agents.

Goal: Create an orchestrator that delegates to researcher, fact_checker, and critic agents.

research_orchestrator (root)
├── researcher     (AgentTool) — web research via Google Search
├── fact_checker   (AgentTool) — independent claim verification
└── critic         (AgentTool) — identifies weaknesses and gaps

Exercise 4: Agent Evaluation

Use ADK's evaluation framework to measure how prompt engineering affects agent quality. Run the same eval suite against a flawed agent and a fixed agent.

Goal: Run adk eval, interpret rubric-based metrics, and see how prompt changes improve scores.

See solutions/exercise_4/INSTRUCTIONS.md for detailed instructions.

Resources

• ADK Documentation
• ADK Python GitHub
• Gemini API

Troubleshooting

Issue	Solution
Agent returns None	Ensure root_agent is defined (not None) in agent.py
Tools not working	Check that tool functions have proper docstrings and type hints
Deployment fails	Check GitHub Actions logs; ensure branch matches workshop/*
API key errors	Verify GOOGLE_API_KEY is set in .env

Need Help?

Ask your workshop facilitator or check the solutions/ folder for reference implementations.