Handling AI API 429 Errors with Retry-After and Backoff in Python

A burst of 50 AI requests hitting a rate limit can trigger a retry storm if every worker retries at the same fixed interval. The article explains that HTTP 429 signals Too Many Requests but doesn't specify which limit was hit, and some providers use it for exhausted quotas, so clients should check documented error codes and headers rather than guess. It recommends distinguishing temporary rate limits from hard quota exhaustion and unknown cases, and handling other errors like 401 or 403 separately. The Retry-After header, in either seconds or HTTP-date format, should be treated as a minimum wait, with exponential backoff and random jitter added to spread retries. A Python helper function demonstrates computing these delays without making network requests.