"Just trying an AI agent" is the most dangerous approach. How to choose the "first task" that only those who achieve results ...