中

Claude Code's New Routine: Your AI Assistant Now Works 24/7

Anthropic, AI Programming, Claude Code, Automation, Routine —

Claude Code shipped a new feature called Routine.

Simple name, practical functionality.

In essence: you define a set of “routine tasks” for Claude Code, and it executes them automatically without you standing over it.

Three trigger modes:

  • Scheduled: Run code checks automatically every morning at 9 AM
  • API-triggered: Kick off tasks via HTTP requests for external system integration
  • GitHub-triggered: Run tests automatically after someone merges a PR

Usage is straightforward. Write a prompt defining what this Routine should do, then set the trigger condition in the right place.

Here’s an example.

I set up a Routine: run automatically at 3 AM daily. “Check if there are any obvious bugs in last night’s commits, generate a brief report and send it to Slack.”

Previously, this kind of work required either my direct involvement, writing scripts, or setting up CI/CD pipelines. Now, you just describe what you need in natural language, and Claude Code handles execution.

Starting to feel like a “cloud employee,” right?

What’s the Real Significance?

After some thought, I realize the interesting part isn’t “automation” itself—it’s that Routine transforms Claude Code from a passive “you ask, it answers” tool into something that actively executes tasks.

Before, Claude Code waited for you. Open terminal, ask a question, get an answer.

Now, you can let it “watch” your project and handle things proactively.

Analogy: before, Claude Code was a dedicated consultant—you call, it comes. After Routine, think of it as a “resident engineer”—it’s right there, handles things independently when issues come up.

This is actually a core trait of AI Agents: proactiveness.

“24/7 Cloud Employee”—What’s the Reality?

Let’s stay grounded and discuss limitations.

Claude Code’s Routine still runs on prompt execution—the quality and outcome depend entirely on how clearly you wrote your prompt.

For well-defined, boundary-clear tasks like “run tests whenever someone merges a PR,” this works perfectly.

But for complex tasks requiring judgment, like “review whether this PR’s architecture is reasonable”—Routine’s effectiveness is still questionable.

Then there’s cost. Every Routine execution consumes tokens. This differs fundamentally from hiring a real “cloud employee” with a monthly salary. Claude Code bills by usage.

If your team runs Routine a hundred times daily, the monthly bill might give you heart palpitations.

My Take

Routine is a solid start. It gives Claude Code genuine “automated execution” capability, not just “answering questions.”

But it’s still distant from being a true “AI employee.”

If you’re interested in Claude Code’s Routine, I’d suggest starting with simple scenarios—code formatting checks, document update notifications—get clear on what it can and can’t do, then gradually expand usage.

Whatever you do, don’t throw every “high-frequency, low-value” task at Claude Code on day one. Not unless you’re prepared for the bill that comes with an “AI worker.”

Reflection Question

What “busy work” would you most want AI programming tools to handle for you automatically? Code review? Testing? Something else?

Related Posts