Isn't this research applicable to today's AI agents?