From Prompt → Thought → Action → Result

AI is evolving fast—and it’s no longer limited to the classic “you ask, it answers” pattern.
Today’s AI can do something far more powerful: act like an intelligent problem-solver.

Modern AI agents don’t just generate text.

They can think through a problem, decide what tools to use, take meaningful actions, and learn from the results before giving you a final answer.

It’s like upgrading from a calculator… to a smart assistant that can research, reason, and execute tasks on your behalf.

In this guide, you’ll get hands-on and build your first real AI agent using Google’s Agent Development Kit (ADK)—the same framework featured in the official Kaggle 5-Day Agents Course.

No prior experience required. Just curiosity, a few lines of code, and you’re ready to create an AI that doesn’t just respond…
it gets things done. 🚀

By the end, you will:

• ✔️ Install and configure ADK
• ✔️ Set up Gemini API authentication
• ✔️ Build your first agent with reasoning + tools
• ✔️ Watch your agent perform an action (Google Search!)
• ✔️ Explore the ADK Web Interface

Let’s begin. 🚀

⚙️ 1. Setup

1.1 Install Dependencies

If you’re using Kaggle Notebooks, google-adk is already installed. To install it locally:

pip install google-adk

1.2 Configure Your Gemini API Key

ADK uses Gemini models, which require authentication.

1) Create an API Key

Go to Google AI Studio → Generate an API key.

2) Add your key to your environment

Inside your project, create a .env file:

GOOGLE_API_KEY="YOUR_KEY_HERE"

1.3 Import ADK Components

These components give your agent LLM reasoning, tools, and execution runtime.

from google.adk.agents import Agent
from google.adk.models.google_llm import Gemini
from google.adk.runners import InMemoryRunner
from google.adk.tools import google_search
from google.genai import types

print("✅ ADK components imported successfully.")

1.4 Configure Retry Behavior

LLMs sometimes hit rate limits or temporary service issues. Retries help make your agent more reliable.

retry_config = types.HttpRetryOptions(
    attempts=5,            # Max retries
    exp_base=7,            # Exponential backoff multiplier
    initial_delay=1,       # Delay before first retry
    http_status_codes=[429, 500, 503, 504]  # Retryable HTTP error codes
)

🤖 2. Build Your First AI Agent

2.1 What Is an AI Agent?

A normal LLM:

Prompt → LLM → Text

An AI agent:

Prompt → Agent → Thought → Action → Observation → Final Answer

Your agent will be able to:

• Interpret a question
• Decide whether to search the web
• Use Google Search
• Combine results into a final answer

2.2 Define Your Agent

Here we describe:

• Name & description
• Model (Gemini 2.5 Flash Lite)
• Instructions (how the agent should behave)
• Tools (the actions it can take — we give it Google Search)

root_agent = Agent(
    name="helpful_assistant",
    model=Gemini(
        model="gemini-2.5-flash-lite",
        retry_options=retry_config
    ),
    description="A simple agent that can answer general questions.",
    instruction="You are a helpful assistant. Use Google Search for current info or if unsure.",
    tools=[google_search],
)

print("✅ Root Agent defined.")

2.3 Run Your Agent

To run an agent, you need a Runner. This acts as the orchestrator that:

• Keeps track of the conversation
• Sends messages to the agent
• Handles tool calls & results

a) Create the Runner

runner = InMemoryRunner(agent=root_agent)
print("✅ Runner created.")

b) Ask the Agent a Question

response = await runner.run_debug(
    "What is Agent Development Kit from Google? What languages is the SDK available in?"
)

This method:

• Creates a temporary session
• Allows the agent to think
• Lets it call Google Search
• Returns the final answer

Perfect for prototyping!

2.4 What Just Happened?

Your agent:

1. Read your question
2. Realized it needed current information
3. Called the Google Search tool
4. Reviewed results
5. Generated the final answer

This is the foundation of AI agents: LLM → reasoning → tool choice → action → observation → improved answer

To see a full trace (thoughts, actions, observations), use the ADK Web UI next.

💻 3. Explore the ADK Web Interface

ADK includes a browser-based UI for:

• Debugging agents
• Viewing traces
• Inspecting tool calls
• Exploring sessions

3.1 Create a Sample Agent

Run:

adk create sample-agent --model gemini-2.5-flash-lite --api_key $GOOGLE_API_KEY

This generates:

sample-agent/
├── agent.py
├── .env
└── __init__.py

3.2 Get Your Proxy URL (Kaggle only)

url_prefix = get_adk_proxy_url()

3.3 Launch the Web UI

adk web --url_prefix {url_prefix}

📌 Important: Do NOT share your proxy link — it contains an authentication token.

🎉 Congratulations!

You just created your first AI agent using Google’s ADK!

Key takeaways:

• Agents ≠ plain LLMs
• Agents reason, use tools, and act
• ADK makes building agents simple and structured
• You now have a working agent that can search the internet

Next steps:

• Try adding more tools
• Build multi-agent systems
• Explore advanced session management
• Deploy agents in real applications