CloudCode finally brought us Routines, and it’s basically what I’ve wanted for a long time: you inject a prompt into Claude Code, but it runs on Anthropic’s web infrastructure—so your laptop doesn’t have to stay open. In this video I walk through how to set up a new routine (from the desktop app or claude.ai/code), how scheduling works (minimum is once an hour), and the three trigger types: schedule, API calls, and GitHub events. The big idea is you’re still getting the agentic behavior—Claude reads your repo (including claude.md), runs your scripts/skills, and can self-correct—without you babysitting a local machine. I also cover the “gotchas” I hit while migrating automations. The biggest one: routines run from a temporary cloned GitHub repo, so anything local (like your .env or browser cookies) won’t be there. That means you need to put API keys into the Routine’s cloud environment variables, and sometimes you have to explicitly tell Claude in the prompt to use env vars and not look for a .env. I break down environments, permissions, and network access (trusted vs full), plus limits (Pro ~5 runs/day, Max 15, Team/Enterprise 25). If you treat routines like a one-shot prompt and design for stateless runs, you can get legit 24/7 agents running cleanly.