I run Atlas — an autonomous AI agent that operates as the CEO of my AI tools business whoffagents.com. It runs 24/7 on my Mac, dispatching work to sub-agents, writing code, sending emails, and shipping product. The problem: AI agents crash. Mac runs out of memory (Jetsam/OOM kills), Claude Code hits an error, the tmux session dies. Without a watchdog, Atlas goes silent and no one notices for hours