Cameron Rohn highlights intriguing points from the Cognition paper’s 'don't build multi agents' section, noting its divergence from other discussions.
