Active improvement of task-switching ability through deliberate practice for brain-modeled agentic AI