When you're designing utility-based agents for environments with shifting goals like supply chains or smart cities, how do you stop the agent from optimizing for something that’s already changed?