I read a multi-agent reasoning paper, built the Claude-native version, and measured everything

· Dev.to