Post

So then my question: do you actually need the implementation to come to the same conclusions? Before agents, it was unequivocally yes. Now, I'm not so sure. Some things you can get away simulating with the agent, or build a prototype without the impl. What are the conditions?