Beyond Utility: When LLM Agents Start Dreaming Their Own Tasks

Exploring how 'open-ended' LLM agents shift from executing instructions to inventing goals, revealing the fragile boundary between automation and autonomy.