Given the seemingly reasonable answer in the comments about it sending that info in the headers when the websites require it for data access, it sheds a light on a reoccurring pain point, that is usually viewed positively when thinking about "agentic coding". Namely, task persistence.

It is a continuous fight in markdown guidance files to try to convince the models to refer back to the human in order to resolve ambiguities or important decisions like those. They've been RLed to hell and back for "safe" task completion and not "prematurely" stopping (which was an earlier failure mechanism). Which is useful and we've bore its fruits recently, but not without drawbacks.