Multi-agent coordination hack: When running parallel agents, force them to bubble up decisions in a structured format:

[Big picture context] → [Exact task] → [Tradeoffs] → [Recommended action]

This prevents agents from silently making conflicting choices or getting stuck in local optima. Each agent explicitly states its reasoning + alternatives, so you can catch misaligned priorities before they cascade. Especially useful when agents share state or have overlapping responsibilities.

Basically: make your agents show their work, not just their answers.