Treat Per-Task Model Switching as a Concurrency Protocol

This article dives into the intricacies of model switching in AI tasks, emphasizing it's more complex than a simple setting change. It's treated as a distributed operation with steps like reading current tasks, preparing configurations, and managing overlapping requests. This complexity means that the system needs a protocol to determine which switch takes precedence when they overlap, which is crucial for maintaining task consistency. MonkeyCode's commit c58bcd4 introduces a method to record these attempts, which is a significant step towards better handling of concurrent model switches in AI systems.