Claude Code's New Routines: Automation Without

Anthropic just rolled out a feature that AI enthusiasts have been jury-rigging for months: scheduled tasks that don't require you to leave your laptop running. The new Routines capability lets Claude Code execute automations in the cloud—daily GitHub scrapes, recurring data pulls, event-triggered workflows—without the old ritual of keeping a terminal window open like some kind of digital hearth fire.

It's a straightforward addition that addresses a genuinely annoying problem. And like most straightforward additions to AI tools, it comes with restrictions that reveal exactly what Anthropic thinks you should be doing with this thing.

The Pitch: Set It and Forget It (Sort Of)

The mechanics are simple enough. You can trigger Claude Code routines three ways: on a schedule (daily at 9 AM, classic cron job style), via API call (on-demand execution), or through GitHub events (when something happens in your repo). Everything runs on Anthropic's infrastructure, deposits results into a GitHub repository you specify, and requires zero babysitting.

Chase AI, who demoed the feature, put it plainly: "I'm sure we've all been in situations where like, wow, I wish I could have Claude just do this one thing every single day and I didn't have to have that exact terminal up. And I also didn't have to create a web app, go host it on Railway, and pay for API fees."

That's the value proposition in a nutshell. Before Routines, your options were: keep a session running (impractical), build and host your own automation infrastructure (expensive and complicated), or just do the task manually every day (defeats the purpose of having an AI coding assistant). Now there's a fourth option that lives in the gap between those extremes.

The Limit: 15 Runs Per Day

Here's where we get to the interesting part. Max-tier users get 15 routine runs every 24 hours. Not 150. Not unlimited with reasonable rate limits. Fifteen.

This tells you everything about how Anthropic is positioning this feature. As Chase noted: "Don't think of this as some replacement for, you know, what you've created in N8N in the past where you're running hundreds of automations in a single day. This is more small-scale stuff for a single user where you want it to run when you're not at your computer and you don't want to have to pay for API costs."

That's a diplomatic way of saying: this isn't automation infrastructure, it's automation convenience. It's designed for the person who wants Claude to pull their morning GitHub trending list or compile a daily summary of project updates—not for the person trying to build a business process on top of it.

Which is probably wise. The last thing Anthropic needs is people using Routines as free compute for production workloads, then complaining when the service gets throttled or restructured. The 15-run limit sets expectations before anyone gets creative ideas.

What It Actually Looks Like

The demo Chase walked through is telling in its ordinariness. He set up a routine to scrape GitHub's top trending AI repositories—top 10 from the last week, top 5 from the last month—and have Claude write a markdown file with an "editor's take" summary.

This is the kind of task that sits in the automation sweet spot: valuable enough that you'd like it done regularly, simple enough that you don't want to build infrastructure for it, and consistent enough that you can trust an AI to handle it unsupervised.

The results? He got exactly what he asked for, deposited in his GitHub repo, with Claude adding a bit of editorial context that his previous Windows-based automation didn't provide. It worked. It saved time. It didn't require him to think about whether his laptop was on.

That's... pretty good, actually. The bar for useful automation is often lower than we pretend.

The Integration Tax

Of course, there are setup requirements. You need a GitHub repository for Claude to dump results into. You need the Claude GitHub app installed if you're using webhooks. You need the GitHub integration connected through Claude's settings for basic scheduled routines. You need to configure your cloud environment, which might already be done if you're on the Ultra plan, but might not be if you're not.

None of this is particularly onerous if you're already in the Claude ecosystem and comfortable with GitHub. But it's worth noting that "automated" doesn't mean "no configuration." You're still connecting services, authorizing access, and setting up infrastructure—just less infrastructure than you'd need without Routines.

Chase ran into the predictable GitHub authorization hiccup during his demo, fixed it in settings, and moved on. This will be someone's debugging session at 11 PM when they're trying to set up a routine and can't figure out why it's failing. Documentation helps, but integration friction is integration friction.

API and Event Triggers: The Other Two-Thirds

Scheduled tasks are the obvious use case, but Routines also supports API triggers and GitHub event responses. Both require web UI configuration—you can't set them up through the CLI, which is a curious limitation but probably reflects some backend complexity.

API triggers seem purpose-built for people who want Claude to respond to external systems, though the 15-daily-run limit makes this feel like a debug-and-prototype feature rather than something you'd rely on for production integrations. "Hey, what would happen if Claude parsed this webhook payload?" is a question you can answer. "Let's route our customer support tickets through Claude" is probably not the intended use case.

GitHub event triggers are more interesting for developers actively working in repositories. Want Claude to automatically analyze pull requests, summarize issues, or track specific repo activity? That's the use case. Whether it's better than just... looking at your notifications is an open question.

What This Actually Changes

Routines doesn't transform what Claude Code can do—it transforms when and where Claude Code can do it. That's meaningful if you're already using Claude for tasks that benefit from regular execution. It's less meaningful if you're trying to decide whether to use Claude at all.

The real question is whether 15 runs per day is enough for the workflows people actually want. For some use cases—morning summaries, daily data pulls, overnight processing tasks—it's plenty. For others, it's a ceiling that'll feel restrictive within a week.

Anthropic will probably adjust this limit based on usage patterns and infrastructure costs. They always do. Whether it goes up or comes with tiered pricing is anyone's guess, but the pattern is familiar: launch conservatively, see what people build, adjust accordingly.

For now, Routines is what it claims to be: a way to automate small-scale, recurring tasks without keeping your laptop running. That's not revolutionary, but it doesn't need to be. Sometimes the most useful features are the ones that just eliminate a specific annoyance.

The question isn't whether Routines works—Chase's demo shows it does. The question is whether your automation needs fit within the constraints Anthropic has decided are reasonable. For some of you, absolutely. For others, you're still going to need that Railway deployment.

— Mike Sullivan, Technology Correspondent