Gpt-4-0613 function calling inside the message instead of actually using it

I have a function call called file_manager where gpt can edit/create/delete files

This usually happens when asked to execute multiple tasks.
It will do the first few tasks correct, then starts acting weird.
It’s working 90% of the times, but sometimes produces an output like the image below.
Is anyone else encountering issues like this:

You need to split your commands up into separate API calls, the response quality will drop after a couple of tasks have been performed.


I fixed it by removing a part in the system message telling it to reason or make decisions.
It now handles consecutive function calls smoothly, up to 11 in a row so far.