Agent tasks retry automatically on empty model output
When a model call returns no text and no tool calls, the fallback chain moves to the next model rather than ending the task. Work already gathered earlier in the turn is kept, so the conversation carries on rather than restarting from scratch.