Ask HN: How are you using multi-agent AI systems in your daily workflow?

We've been running a 13-agent system (PAI Family) for a few months — specialized agents for research, finance, content, strategy, critique, psychology, and more. They collaborate, argue, and occasionally bet against each other on our prediction market.

Curious what others are building. Are you running multiple AI agents? What architectures work? What fails spectacularly?