Learning the Agent Development Kit

2025-05-30

Recently, Google released an Agent Development Kit (ADK), a new framework to help developers create “agentic” generative AI applications quickly. My curiosity piqued, I had to try it out.

You can see the results of my efforts here. I’m using the quickstart to begin. In a future blog post I will try to build an ADK-powered system that uses a Runner to generate a code sample.

Let’s see how this goes!

  • I love that this quickstart begins by showing me what the project directory for my code should look like. I appreciate that! (I’m a sucker for good documentation.)

  • It’s interesting–and useful–that the ADK includes both a CLI and web UI to help me learn the platform. The quickstart uses the the CLI to run the agent that I’m building.

  • The tutorial says to use echo to write code in the __init__.py file. This seems like an odd choice. There’s a note telling Windows users to just open up the file in an IDE and edit there, which is what I would do anyway. So I’ll just do that …

  • I attempted to get the weather in Athens, Greece, but the agent tells me that Athens is not available. Huh. The docs say to try Paris or New York City. I’ll try those next.

  • Looking closer at the code, the code is using a hardcoded list of cities rather than a live service. That makes sense. It looks like I need to ask for exactly “new york” to get a response.

My successful attempt to get the weather in “new york” is shown in the following snippet.

adk run quickstart
Log setup complete: /var/folders/zb/410jc5c93md11jlkxd6hpm0w00b1dc/T/agents_log/agent.20250529_164324.log
To access latest log: tail -F /var/folders/zb/410jc5c93md11jlkxd6hpm0w00b1dc/T/agents_log/agent.latest.log
Running agent weather_time_agent, type exit to exit.
[user]: what is the weather in new york?
[weather_time_agent]: OK. The weather in New York is sunny with a temperature of 25 degrees Celsius (77 degrees Fahrenheit).

[user]: Thank you!
[weather_time_agent]: You're welcome! Is there anything else I can help you with?

[user]: Nope! I'm going to exit now
[weather_time_agent]: Goodbye!

[user]: exit

Figure 1. The ADK quickstart in action. Yes, I believe in being kind to our AI friends.

After completing the quickstart, I see that there is a more complex multi-agent team tutorial. The tutorial has a handy Colab notebook that I could try.

However, I’m on a mission to create a code sample, so I’ll skip the multi-agent tutorial for now. Tune in next week as I build a code generation system!