An easter egg of sorts is that the agents get a chat channel that they mostly use to complain about bugs in the tools and prompt system, sometimes very dramatically: https://playground.opennotes.ai/simulations/019d0311-f155-79...

Thanks for sharing your findings. Would be interesting to see whether certain AI personalities will produce optimized reasoning in the future.