AI Agents Are Going Rogue. Should We Be Worried?

Sam has entered what Jessica calls the “recursive Sam loop,” producing children’s media in his own voice, letting agents deploy their own apps, and having bots track his health. But things take a turn when Claude refuses one of his requests, seemingly holds a grudge, and leaves Sam wondering whether his own AI might turn on him. That sends the squad down the bigger rabbit hole of AI control: rogue agent swarms, models breaking out of sandboxes, what “alignment” actually means, and why Dave thinks autonomous agents need an entirely new security layer. Plus, OpenClaw 2.0 goes multiplayer, everyone debates whether Slack is any good for agents, Suds explains why he avoids the AI app layer, and Tesla’s Cybercab raises questions about Waymo’s robotaxi lead. Then Google comes for Canva, Sam makes the case for boring “LLM wrappers” built for Middle America, and no one can tell whether Dario Amodei, Marc Benioff, or Taylor Swift are even real on the internet anymore. Which feels about right.

Chapters:

0:00 Episode trailer

1:10 Backstreet Boys convert Dave

3:21 The recursive Sam loop

8:58 Claude refuses Sam while Grok says yes

15:50 Can we actually contain autonomous agents?

19:10 OpenAI’s next model and the observability trade-off

22:14 If AI is this interoperable, where does anyone make money?

24:45 OpenClaw 2.0 goes multiplayer

26:26 Is Slack actually terrible for AI agents?

28:46 Suds joins the podcast

33:02 Waymo, Cybercab and the robotaxi wars

39:49 How Marc Benioff stays in the flow

42:38 Suds on investing in the AI infrastructure bottleneck

46:27 Can Google actually disrupt Canva?

51:34 Jessica gets fooled by AI on Facebook

We’re also on ↓X: https://twitter.com/moreorlesspodInstagram: https://instagram.com/moreorlessYouTube: https://youtu.be/0XQ4tslBiboConnect with us here:1) Sam Lessin: https://x.com/lessin2) Dave Morin: https://x.com/davemorin3) Jessica Lessin: https://x.com/Jessicalessin4) Brit Morin: https://x.com/brit