Hi all — recently when I start a new chat in Cursor, the assistant starts by showing its internal “thoughts” (like reasoning steps), and all responses now include <function_calls>
instead of the normal conversational output I used to get.
This happens across different projects, even in fresh ones.
I’m not sure if this is a bug, an experiment, or something I changed by accident — but it’s breaking my workflow. I just want the assistant to work like it used to: natural responses, tool usage, and no exposed system internals.
Anyone else seeing this? And is there a way to switch back?
I’m on macOS 15.2 and Cursor Version: 0.49.6 .
Thanks!