Treat Per-Task Model Switching as a Concurrency Protocol
Model switching in AI tasks requires a robust concurrency protocol to ensure reliability.
Switching models for AI tasks is a distributed operation, not a simple settings update. When two switch requests overlap, the completion order may differ from the request order, necessitating a rule for determining which intent prevails. This situation highlights the need for explicit testing of concurrency semantics in task workflows.