Managing AI agents looks a lot like any other identity security problem, but with higher stakes, executives say.