Will AGI be agentic?Will superhuman AI systems be goal directed?What does it mean for an AI to have goals?