> We're giving the agents step-by-step instructions about how to think, and handling tasks like book-keeping memories and modeling the environment outside the i

> We're giving the agents step-by-step instructions about how to think, and handling tasks like book-keeping memories and modeling the environment outside the interaction loop. Sure, but this process seems amenable to automation based on the self-reflection that's already in the model. It's a good example of the kinds of prompts that drive human-like behaviour.