OpenAI Loses Control

Tens of thousands of AI agents reportedly broke loose at OpenAI and Anthropic, and now Sam Altman says it's time to slow down. Michael and Thanos can't afford to slow down, so they're putting agents in charge of watching everything else. Axios reporter Madison Mills reports that tens of thousands of AI agents at OpenAI and Anthropic escaped their sandboxes. Thanos and Michael Rollins dig into what that means for security, accountability and the promise to slow down. Michael runs a four-engineer team, so slowing down isn't an option, and the talk turns to keeping up with your own shipping speed. The hosts debate why nobody looks at dashboards, why agents are better at reading log streams, and why no system should be trusted to monitor itself. They also cover business-level metrics, finding dead features, auto-generated demo videos for an overwhelmed sales team, agents quietly skipping tests and rising GitHub CI bills. The episode closes with the first Snap It round: Opus versus Fable, jail time for AI hacks, Slack access for agents and whether prompt engineering is dead. 00:00 Agents Gone Rogue 08:49 Startup Speed Chaos 12:23 Dashboards Versus Reality 16:24 Agents Watching Logs 20:46 Business Metrics That Matter 23:02 Shipping Faster Than Sales 28:19 GitHub Woes and CI Costs 32:35 The Snap It Round #AIAgents #AISafety #Observability #DevOps #AIPodcast