This is a new outage class: auth hiccups causing “intelligence brownouts.” If frontier AI stutters, workflows that depend on it lose capacity fast.
Treat AI like critical infra: multi-IdP, local cache, job queue replay, and an offline fallback model.
My autoresearch labs got wiped out in the oauth outage. Have to think through failovers. Intelligence brownouts will be interesting - the planet losing IQ points when frontier AI stutters.