The world of AI agents is moving at lightning speed. Keeping up with the latest trends, new open-source tools, and important developer conversations can feel like a full-time job. We're facing this exact challenge as we prepare to launch our new podcast for agent developers, which will be dedicated to all things AI agents (stay tuned!). To make sure we're ready for each episode, we wanted to create an automated way to get up-to-date with the news that matters.
In this guide, we are going to walk you step-by-step through building your very first AI agent on Google Cloud using the open-source Agent Development Kit (ADK). We will design a "Trend Spotter" agent whose mission is to act as your personal AI analyst, teaching it to scan the web and sift through the noise to find what truly matters.
By the end of this post, you will have a practical, working tool that automatically creates a concise intelligence report to keep you up-to-date, saving you hours of manual research. More importantly, you will learn the fundamental skills to build your own agents with ADK. You will know how to: Structure a simple, powerful ADK agent as a proper Python package. Write a detailed prompt to define your agent's logic and workflow. Provide your agent with tools like Google Search. Set up, test, and run your agent locally using the adk web interface. Deploy your agent to Cloud Run.
This setup uses a standard package structure that allows the ADK tools to discover and run our agent without a main.py file.
Step 1.1: Prerequisites Python 3.11+ Google Cloud CLI: Follow the official installation guide here. Google Cloud Account.
Open your terminal. Create the following folder structure and virtual environment.
These settings tell ADK how to securely connect to your Google Cloud account to use services like Vertex AI and Google Search. Set Environment Variables: In your terminal, run the following export commands. These tell ADK to use the Vertex AI platform in your specific Google Cloud project and region. Log In to Your Account: Run this one-time command. It will open a browser for you to sign in, allowing ADK to make authorized requests on your behalf.
Now we'll write the code and place it inside our trendspotter package directory.
Note that we are guiding the LLM to specify the date range in the call to the GoogleSearch tool to make sure we are focusing on trends from the last week.
To use the adk web command, we need to tell ADK where to find our agent. We do this in the pyproject.toml file.
Run this command from your project's root directory. The -e . command installs your project in "editable" mode so the adk tool can find it.
Open the URL that appears in your terminal. In the web interface, select "trendspotter" from the dropdown menu. You can now chat with your agent! Ask it: "Generate a report on the latest AI agent news." This might take a few minutes, depending on the amount of searches you instruct the agent to perform in your prompt.
The adk web interface is your best debugging tool. On the "Events" tab, you can see every step your agent takes, including which tools it calls and what the LLM is thinking. If the output isn't right, your first step should always be to adjust the instructions in prompt.py.
Ensure you have authenticated with Google Cloud (gcloud auth login and gcloud config set project ) and setup your environment variables to deploy your agent to cloud run with one line command.
Optional but recommended: Setting environment variables can make the deployment commands cleaner.
You can test your agent by simply navigating to the Cloud Run service URL provided after deployment in your web browser. (The URL should be similar to this: https://your-service-name-abc123xyz.a.run.app)
Congratulations! You have successfully designed, built, tested and deployed your very first AI agent using the Agent Development Kit.
You've learned how to structure a proper agent package, how to write a detailed prompt to control an agent's logic, and how to run and interact with your agent using the adk web interface. We now have a working "researcher" for our AI agent podcast and you you now have a working foundation that you can expand upon. Try modifying the prompt to research a different topic, or explore adding new custom tools to give your agent more capabilities.