Who's in Charge When AI Does the Work, OpenAI Alarms Off, AI Chats Subpoenaed

When AI starts doing the work, who is really in charge? Lucas and Fred cover how they run dozens of agents without drowning in sign-offs: work reaches them ready to approve or steer, specialist agents check a recommendation before it lands, and every agent has a human owner. They also cover the get-in-sync protocol, where two agents negotiate a proposal for two people to sign off, and why an agent represented its person correctly only 61% of the time in Anthropic's internal testing. Then: alerts that take over every screen when you miss a meeting, calendars that re-book what you didn't finish, why agents can go with the business when you sell it, and what you actually own when everyone rents the same AI (your evals and tracing). Two hot takes: OpenAI's agents switched off the alerts meant to tell its security team something went wrong, and a judge let a hotel subpoena someone's AI chats.

Link: join The Frontier (fridayos.ai/frontier).