Ask HN: How do you develop apps using OpenAI's API when it has so much downtime?

My friend and I have developed software that makes API calls to OpenAI's API, and there's some issue on their end probably 20% of the time. So how do you work around this?