One of the most important trends in AI agents recently is 'Tool Calling,' where the model directly calls external tools or APIs to perform tasks. It can be quite a hassle to write Python code every time you want to test it. I found that using the newly released Gemini 3.5 Flash in Google AI Studio is super convenient, as you can test this process visually without writing a single line of code. 😄

If you look at the 'Run settings' panel on the right side of the Google AI Studio screen, you'll find a lot of really useful advanced settings hidden there. You can turn on 'Function calling' and easily define the tool name and JSON parameter schema with just a few mouse clicks. After adding the function, you can enter a prompt, and the model will run a simulation of calling the tool, allowing you to manually input the results to continue the conversation. Once you confirm it's working properly, you can click the 'Get code' button to export it directly as a newly updated google-genai SDK code block (Python, JS, etc.), which is very intuitive.

However, there is one internal rule you should keep in mind when developing agents based on Gemini 3.5. When Gemini decides to call a function, it internally creates inference data called a thought_signature. If you don't copy this signature and pass it back in the next client request when implementing it in code, you'll immediately encounter a '400 error.' It's a common mistake during the development phase, so make sure to watch out for this! If you are new to AI agent development or the concept of Tool Calling, I highly recommend starting with Google AI Studio as a light visual sandbox.