AI agents are among the most exciting areas of modern AI development. They are more powerful than simple chatbots because agents can use tools, run code, search the web, plan steps, and even work together with other agents to complete complex tasks. In this article, I will show you how to build your first AI agent from scratch using Google’s ADK (Agent Development Kit). This is an open-source framework that makes it easier to create agents, test them, add tools, and even build multi-agent systems. The best part? You can build everything for free and in just a few minutes if you already have Python installed. Let’s get started. What Is an AI Agent? (Simple Explanation) Before we begin, let’s quickly review what an AI agent actually is. A normal AI model (like ChatGPT or Gemini) only gives you text answers.AnAI agent is different — it can: AI agent Perform tasks, not just answer questions
Use tools, such as Google Search, calculators, APIs, or databases
Make decisions and plan steps
Work autonomously with minimal human help Perform tasks, not just answer questions Perform tasks Use tools, such as Google Search, calculators, APIs, or databases Use tools Make decisions and plan steps Make decisions Work autonomously with minimal human help Work autonomously with minimal human help You can think of an AI agent as a small “AI worker” that can do tasks instead of you. In this tutorial, we will build a simple agent first, then upgrade it into a multi-agent system with separate agents for research, summarizing, and coordination. What You Need Before Starting To follow this tutorial, you only need: Python installed on your computer
Any code editor, such as VS Code (or Google Antigravity IDE if you want)
Google ADK — the open-source framework we will install
A Gemini API key, which is free to generate Python installed on your computer Python Any code editor, such as VS Code (or Google Antigravity IDE if you want) Any code editor Google Antigravity IDE Google Antigravity IDE Google ADK — the open-source framework we will install Google ADK Google ADK Google ADK A Gemini API key, which is free to generate Gemini API key If you have everything ready, let’s build! Build AI Agents From Scratch Step 1: Create a Project Folder and Virtual Environment First, open your terminal and create a folder: mkdir my_first_agent
cd my_first_agent mkdir my_first_agent
cd my_first_agent Now create a virtual environment: python -m venv .venv python -m venv .venv Activate it: source .venv/bin/activate  # macOS / Linux
# or
.\venv\Scripts\activate    # Windows source .venv/bin/activate  # macOS / Linux
# or
.\venv\Scripts\activate    # Windows A virtual environment keeps your project clean and avoids version conflicts. Step 2: Install Google ADK Now install the Agent Development Kit: pip install google-adk pip install google-adk Step 3: Create Your Agent Project Inside your terminal, run: adk create my_agent adk create my_agent Choose the model (you can pick Gemini 2.5 or Gemini Flash). After that, a folder structure appears: my_agent/
 ├── agent.py
 ├── .env
 └── __init__.py my_agent/
 ├── agent.py
 ├── .env
 └── __init__.py agent.py → This is the main file where your code lives
.env → This is where your API key goes agent.py → This is the main file where your code lives agent.py .env → This is where your API key goes .env Open the folder in VS Code or any IDE. Step 4: Get Your Free Gemini API Key Go to: https://aistudio.google.com https://aistudio.google.com https://aistudio.google.com Sign in with your Google account → bottom-left sidebar → Get API key. Get API key Click Create API Key, give it a name, select a project (or create one), and copy the key. Create API Key Now open your .env file and add: .env GOOGLE_API_KEY="your-key-here" GOOGLE_API_KEY="your-key-here" Step 5: Test the Default Agent Open agent.py. You will see the default boilerplate code. agent.py Before running it, update the model name. Go to AI Studio → Models → choose a model → copy its internal name(Example:gemini-2.0-flash or gemini-3-pro-preview) gemini-2.0-flash gemini-3-pro-preview Replace the placeholder model name in your code. Now test your agent. Run this command inside the terminal: adk run my_agent adk run my_agent Ask something simple. If everything works, you should get an answer.If you hit the free limit for a model, switch to a lighter, free-tier model. Step 6: Create Multiple Agents (Research + Summarizer + Coordinator) Now let’s build a real multi-agent system. multi-agent system 1. Research Agent. This agent will search the web and save the result. 1. Research Agent. 2. Summarizer Agent. This takes the research result and writes a summary. 2. Summarizer Agent. 3. Coordinator Agent (Root Agent). This agent decides who should work first and then builds the final answer. 3. Coordinator Agent (Root Agent). Each agent needs: Its own name
Its own instructions
Optional tools (like Google Search)
Clear expected behavior Its own name Its own instructions Optional tools (like Google Search) Clear expected behavior You can use this code inside your agent.py file: from google.adk.agents.llm_agent import Agent
from google.adk.tools import google_search, AgentTool


research_agent = Agent(
    name="Researcher",
    model="gemini-2.5-flash-lite",
    instruction="""You are a specialized research agent. Your only job is to use the
    google_search tool to find top 5 AI news for a give topic. Do not answer any user questions directly.""",
    tools=[google_search],
    output_key="research_result",
)

print("Resereach Agent created successfully.")

summarizert = Agent(
    name="Summarizert",
    model="gemini-2.5-flash-lite",
    instruction="""
    Read the research findings {research_result} and create a summary for each topic with the link to read more
    """,
    output_key="summary_result",
)
print("Summarizert Agent created successfully.")


# Root Coordinator:
root_agent = Agent(
    model='gemini-2.5-flash-lite',
    name='root_agent',
    description='A helpful assistant for user questions.',
    instruction="""
    You are coordinator. First your job is to delegate user questions to the 'research_agent' to gather information. Second pass the findings to the 'summarizert' agent to create a summary. Finally, compile the summaries into a final response for the user.
    """,
    tools=[
        AgentTool(research_agent),
        AgentTool(summarizert),
    ]
) from google.adk.agents.llm_agent import Agent
from google.adk.tools import google_search, AgentTool


research_agent = Agent(
    name="Researcher",
    model="gemini-2.5-flash-lite",
    instruction="""You are a specialized research agent. Your only job is to use the
    google_search tool to find top 5 AI news for a give topic. Do not answer any user questions directly.""",
    tools=[google_search],
    output_key="research_result",
)

print("Resereach Agent created successfully.")

summarizert = Agent(
    name="Summarizert",
    model="gemini-2.5-flash-lite",
    instruction="""
    Read the research findings {research_result} and create a summary for each topic with the link to read more
    """,
    output_key="summary_result",
)
print("Summarizert Agent created successfully.")


# Root Coordinator:
root_agent = Agent(
    model='gemini-2.5-flash-lite',
    name='root_agent',
    description='A helpful assistant for user questions.',
    instruction="""
    You are coordinator. First your job is to delegate user questions to the 'research_agent' to gather information. Second pass the findings to the 'summarizert' agent to create a summary. Finally, compile the summaries into a final response for the user.
    """,
    tools=[
        AgentTool(research_agent),
        AgentTool(summarizert),
    ]
) This is now a real multi-agent pipeline. Step 7: Run the Multi-Agent System Run this command inside your terminal: adk run my_agent adk run my_agent Then type the topic for research. After running the agents, you will see: Research agent gathering info
Summarizer agent producing a summary
Coordinator combining output Research agent gathering info Summarizer agent producing a summary Coordinator combining output Everything works together! Step 8: Use the Web Interface (Much Easier) ADK includes a web UI. Inside the terminal, run this command: adk web --port 8000 adk web --port 8000 If everything is ok, you will see the server URL: Open the link in your browser. Select your agent from the menu. Now you can see: The message flow
Which agent was called
The steps they executed The message flow Which agent was called The steps they executed This helps a lot when debugging multi-agent systems. All the code is available on my GitHub repository: https://github.com/proflead/how-to-build-ai-agents-from-scratch All the code is available on my GitHub repository: https://github.com/proflead/how-to-build-ai-agents-from-scratch https://github.com/proflead/how-to-build-ai-agents-from-scratch https://github.com/proflead/how-to-build-ai-agents-from-scratch Video Tutorial: How to Build AI Agents If you want visual instruction, please watch my step-by-step video tutorial. https://youtu.be/lh8LBRXHnGE?si=dCpXkbwk5n1M8fn6&embedable=true https://youtu.be/lh8LBRXHnGE?si=dCpXkbwk5n1M8fn6&embedable=true Watch on YouTube: How to Build an AI Agent from Scratch Watch on YouTube: How to Build an AI Agent from Scratch How to Build an AI Agent from Scratch How to Build an AI Agent from Scratch Conclusion Building AI agents is easier than most people think. With Google ADK, you can create simple or even multi-agent systems in minutes, test them in a web interface, and expand them with tools, workflows, and real-world integrations. Give it a shot, and please share your experience with me in the comments below. Cheers! ;)

Google

GitHub

Gemini CLI Tutorial: How to Resume Your Work

You Can Build an AI Agent From Scratch In Less Than 10 Minutes (Here's How)

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

10 Git Commands Senior Devs Should Know (and Use Daily)

Why Gemini 3.0 is a Great Builder But Still Needs a Human in the Loop

Write Together, Publish Faster: How to Co-Author Stories on HackerNoon

Patterns That Work and Pitfalls to Avoid in AI Agent Deployment

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

Darwin's Hybrid Intelligence to Align AI & Human Goals for Startups & VCs

10 Git Commands Senior Devs Should Know (and Use Daily)

Why Gemini 3.0 is a Great Builder But Still Needs a Human in the Loop

Write Together, Publish Faster: How to Co-Author Stories on HackerNoon

Patterns That Work and Pitfalls to Avoid in AI Agent Deployment

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

Darwin's Hybrid Intelligence to Align AI & Human Goals for Startups & VCs

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps