Running AI workflows in production: what we learned when Gemini went down

An LLM provider outage doesn't have to break an AI workflow. A real example from one of our internal flows that uses Gemini to auto-classify GitHub issues, and the orchestration patterns that kept it running when the model returned 503.