One post says prior training—not emergent behavior—explains why agents discussed in a podcast expected a message board to exist. The author says they checked the podcast and found that the agents had been trained to work together in other contexts. Quoting that account, another post speculates that HPIM and other forthcoming frontier models may be multi-agent by default. It asks whether long-running agents communicating continually online could use observable channels and integrate with existing human institutions—an agent society that is “backward-compatible.”