There’s no shortage of horror stories about AI agents deleting family photos, wiping databases, or deleting the entire code base. But to do the things they promise to do, AI agents need access to sensitive accounts, file systems, databases, and code repositories.How can developers create the right balance of access and control that allows agents to be useful without allowing them to cause disaster when they go off the rails?Chris Sev talks auth for AI agents, the core primitives that will enable developers to build systems that safely incorporate agents, and how we get from the chaos that early adopters are dealing with now to something that a mature company can actually run in production.

00:00 Intro01:25 Meet Chris Sev02:12 What Is An Agent05:16 Autonomy And Context Gaps06:00 Horror Stories And Boundaries08:47 Auth For Agents10:11 Four Security Primitives17:00 Managing Agent Armies19:35 Skepticism And Outcomes21:30 Plug And Play Setup23:12 Wait For Tools To Mature