The models that are used for understanding and calling plugins and understanding and calling the python sandbox are trained and prompted differently. For now, they are separate.
It may be a challenge to integrate these two different ways of interfacing with outside code, as each takes special model fine-tuning. The AI may be confused and do the wrong thing.
There is a chance. There’s also reasons why you wouldn’t want an AI that can write and run code to be able to access plugins that can get it to outside services.