AI personal assistant isn't a new thing by any means. There has been Assistant from google or Amazon Echo already available for years. But so far these applications have been rather limited and only capable of doing tasks that have been preprogrammed to them. The biggest thing they brought was combining such task based agents with speech recognition.
The latest boom of LLM based AI applications has emerged a new wave of personal assistant applications. Thanks to the models the assistant are able to "understand" wider variety of requests. They are also able to process users personal data better and understand the actual personal needs beyond the capability of the previous generations ability ro practically just remember the users name and provide "personalized" replies based on that and few other details.
While the field is still new there are also agents that are already capable of executing some of those tasks on their own, even iterating through multiple layers of tasks to be able to reach the desired outcome. Of course this is the part where things can go horribly wrong, so obviously the providers have been careful adding such capabilities to their assistants. So far the most capable assistants are mostly able to process the users personal information and success the course of action based on that data which the user still needs to carry out on their own.