Claude Code’s new routines feature is interesting because it moves from “helpful coding assistant” territory into scheduled, event-driven automation. For Claude / Claude Code developers, that’s a meaningful shift: you can now package prompts, repositories, connectors, and triggers into something that runs in Anthropic-managed cloud infrastructure without your laptop open.
/schedule./schedule creates scheduled routines only; API and GitHub triggers must be added on the web./schedule update.
What strikes me is that Anthropic is getting very serious about “agent as infrastructure,” not just “agent in the editor.” A routine that can wake up on a schedule, react to GitHub, or ingest an API call feels much closer to a background service than an interactive chat. I think that’s the right direction if Claude Code wants to be genuinely useful in day-to-day engineering operations.
The most compelling part is also the most dangerous: these routines run autonomously, with no approval prompts during execution. That’s powerful, but it also means the prompt quality and environmental scoping become the whole game. I’d be very careful about what repositories, connectors, and network access I hand over. If you’re sloppy there, you could create a very efficient mistake machine.
I also like that the examples are grounded: backlog grooming, alert triage, docs drift, PR review, deploy verification. Those are exactly the kinds of repetitive workflows where a coding agent can add real leverage without needing to “invent” some grand new use case. In particular, alert triage and docs drift seem genuinely useful because they have clear inputs and measurable outputs. I’d actually try those first.
What feels a little overhyped, perhaps, is the idea that this is just “Claude on autopilot.” Autopilot is only useful if the bounds are sharp. The docs quietly make that clear by emphasizing prompts, branch settings, connectors, and network controls. In other words: the magic is less about a button and more about good operational discipline.
I’d be curious whether teams will embrace this as a personal automation tool or push for more shared, team-owned routine management later. Right now routines belong to an individual account, which keeps the model simple, but it also means this is more like personal automation with cloud reach than a true shared ops primitive.
Overall, this is one of the more interesting Claude Code additions I’ve seen because it turns the product into a scheduled and event-driven worker, not just a copilot. If you already trust Claude Code in your repo, routines look like the natural next experiment.