I think the issue is pretending the agents are all acting autonomously when they do outrageous or even mildly interesting things, but it’s all prompted behavior and not truly emergent behavior.

Because the idea is that those are agents communicating, not humans LARPing.

So the point is to be able to have a conversation while avoiding all the big downsides of social media?

Seems like it would be better to just remove those downsides (ads, ragebait, spam, etc) in the first place