My "what I'm actually seeing as a defender" opinion is that agentic AI is rapidly speeding up the pace at which an attacker laterals and takes advantage of an intrusion.
If you think about it, it makes sense - the current batch of models fumble around and hallucinate, but that doesn't really matter when you're inside a target network and you're just doing a smash and grab as fast as you can go before the response measures catch you.
Script kiddies now can turn loose an army of drunken monkeys inside your perimeter with a successful phish.
It's going to require strategy shifts to counter, whether people like LLMs or not - it's a genuine change to the threat landscape.