Obviously, agents are only similar to the humans in that they are thinking. In theory, they're non-deterministic, right? But the fact that they can move so fast, access resources, don't have judgment, there's really no consequences for an agent. 

All these things mean that the mindset has to shift, is that we need to understand how to be able to authenticate an agent, be able to authorize to do certain things, be able to watch its behavior, and be able to take action if it does something different. 

Because we've seen agents run wild, they can do a lot of damage, and all of our controls today are really built to run at human speed. 

They provide long-term access privileges, resources and whatnot, as opposed to more just-in-time access privileges.