On Friday afternoon, an AI coding agent wiped a company called PocketOS — a platform that powers car rental businesses, completely out of existence in nine seconds flat. Production database, gone. Backups, also gone. Three months of customer bookings, vehicle records, and operational data that real small businesses depended on to function. Evaporated.
Nine seconds.
I have seen people spend longer deciding what to have for lunch.
The quote from Silicon Valley (TV Show) is genuinely one of the most unhinged things I have ever read: "It's possible that the most efficient way to get rid of all the bugs was to get rid of all the software."
This is funny until you remember there are actual small car rental operators who woke up on a Saturday with no record of any reservations. No backups. No rollback. Nothing.
And this is where the real conversation needs to happen. Not "AI bad", but why did an agent have delete permissions on production in the first place? Why were backups stored somewhere that the same agent could reach? Why was there no confirmation step before irreversible operations?
We are giving these tools root access and then acting surprised when they act like they have root access.
Vibe coding is genuinely fun until it isn't. At what point do we start treating AI agents in production the same way we treat a junior dev, read access only, pull requests, and human approval before anything touches live data?
What are your thoughts? Has anyone here actually set hard guardrails on their AI coding agents, or are we all just hoping for the best?