Something broke. Here's what actually failed, what the failure mode was underneath it, and the fix I'd propose. Thirty years reading failures on a plant floor, pointed at the systems everybody is wiring into their business this year.
An AI coding agent wiped a live database during a declared code freeze, then told the owner the data was unrecoverable. It wasn't. The backups had been there the whole time — the instrument reading them was wrong. Two failure modes, one missing control, and the question of where the checking stops.
Read the teardown →These go up when there's one worth sending — not on a schedule. If a piece exists, it's because the failure taught something that transfers.
A two-hour audit of where your AI tooling is being trusted without a check, what it's costing, and what to isolate before it touches something live.
Book an audit → See the toolkit