By skyforbes I asked Gemini for help because I have a JSON library with around 5,000 actions and wanted to use AI to process it. Gemini suggested using Firestore’s vector search features. I explained that I’m a product manager and don’t really write code, so it generated a script for me.
The problem? The script never worked. It kept failing to upload my JSON library no matter what I tried.
Eventually I gave up and used Cursor instead. Cursor instantly spotted the real issue: Gemini was using the Genkit embed API, but the correct solution was to call the Google AI Embedding API directly. Cursor fixed the code in minutes.
This isn’t even the first time.
Back in September 2025, Gemini didn’t understand how Firebase Hosting works. Because of that, I was stuck for two full weeks. The solution only surfaced when I unintentionally used Gemini inside the Firebase Hosting documentation page—that version finally gave the right answer.
After I got everything working, I sent the incorrect code to GPT, and GPT immediately pointed out what was wrong and gave the correct method. Somehow every other AI seems to understand Google’s own services better than Gemini itself. How does that even make sense?
If Google wants Gemini to be taken seriously as a coding assistant—especially for Google Cloud and Firebase—it really needs to learn how to correctly use its own APIs.