This article explores the challenges of managing AI operations, focusing on the need for task queues to prevent context loss, duplication, and improve observability. It details how queues can be used to manage variable token costs, rate limits, and ensure deterministic operation.