So I use Gemini 2.5 Pro in an app that I developed in AI Studio, but it is now deployed to my own web server. The purpose of the application is to generate new C# projects from scratch, where I have defined several restrictions on what can be included and excluded.
One of the things my app has is a file employee.json which contains demo content with companies and employees. My app is told to always include this file in a new /Shared folder of my C# project. Literally copy and paste.
It fails at this and keeps generating a new file with content, and this content is often way too different from my intended use. I have this specific dataset because it is used in all my projects. It’s just that any new project has to use it for data-seeding when it needs a user database or something like that.
But Gemini is just ignoring my request and keeps generating a garbage JSON file instead of what I need. After ten attempts to tell it to use this specific JSON file, I gave up and added the file manually and told the AI to use the file I just added…
Yep. It replaced the content with garbage again, because the data did not match the data model it asked for…
Edited it again and this time I locked the file in my App. So glad that I added this lock functionality in my App! 
But it is annoying that the API just keeps ignoring what I want and just decides to do what it thinks best.
So the problem is that the API thinks it’s smarter than me, the user, and just does what it wants to, ignoring my request. And this isn’t the first time I had to deal with this. (AI Studio also has this problem.)
So, it would be nice if the API would listen to us users more often instead of doing what it thinks is best. Because the AI tends to do things that we don’t want…