Mail

"I set up automation, but do I need to keep my computer on?"

Anyone who has used no-code automation tools like n8n, Zapier, or Make has asked this at some point. If you run automation on your laptop without a server, that laptop has to act as the server. Close it, and everything stops.

In April 2026, Anthropic released a solution: Claude Code Routines — AI agent automation that runs in the cloud. Even when your computer is off, Anthropic's servers keep the agent running.

As an EdTech CEO who uses Claude Code daily, here is what Routines actually are and how to use them effectively.

What Are Routines — How They Differ from /loop
Three Trigger Types — Scheduled, API, GitHub
Setup Guide — Step by Step
Five Real-World Scenarios
Costs, Limits, and What to Watch Out For

1. What Are Routines — How They Differ from /loop

A Routine is a prompt, one or more repositories, and a set of connectors packaged into a single automated unit.

Claude Code already had /loop for repeated local execution — running the same prompt at a set interval on your own machine. Routines take this significantly further.

Feature	/loop	Routines
Execution location	Local machine	Anthropic cloud
When computer is off	Stops	Continues
Trigger method	Time interval	Schedule / API / GitHub events
Setup complexity	Simple	Moderate
Plan required	Free and above	Pro and above

The core difference: you don't need your computer running. Anthropic's cloud infrastructure executes the agent, so you can configure it once and forget it.

"Using a tool and operating an agent are different things. Routines are the first official gateway to creating agents that work without you."

2. Three Trigger Types

Routines can be started in three ways.

Trigger 1: Scheduled

The simplest option. Routines run on a recurring schedule — hourly, nightly, every Monday at 9 AM, whatever fits your workflow.

Every morning at 6 AM: summarize previous day's GitHub issues → post to Slack
Every Friday: review codebase changes → generate PR draft
First of each month: compile monthly report → write to Notion automatically

Trigger 2: API Call

An external service or your own app sends an HTTP POST request to start the Routine.

For example: when a student submits an assignment → a feedback agent automatically runs.

curl -X POST https://api.anthropic.com/v1/routines/{routine_id}/trigger \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -d '{"input": "Analyze new submission file"}'

Trigger 3: GitHub Event

Reacts to repository events. When a pull request opens, a release is published, or an issue is created — connect the event to a Routine and it runs automatically.

Claude Code Routines trigger types overview

3. Setup Guide — Step by Step

Prerequisites

Claude Code Pro plan or above ($20/month)
Claude Code CLI v2.1.91 or above (v2.1.101+ recommended)
GitHub repository (for GitHub triggers)

Option A: Set Up via Web

Go to claude.ai/code/routines
Click New Routine
Configure name, prompt, repository, and trigger
Save and run a test

Option B: Set Up via CLI

In an active CLI session, type:

/schedule

This connects to the web editor where you can configure the Routine.

4. Five Real-World Scenarios

Scenario 1: Daily Morning Briefing Agent (for Educators)

Every morning, the AI automatically compiles the day's EdTech news, a checklist of class prep items, and outstanding issues — then posts them to Slack. You arrive to a complete briefing.

Trigger: Scheduled daily at 7 AM

Scenario 2: Automatic PR Code Review (for Dev Teams)

Every time a PR is opened on GitHub, Claude automatically reviews the code and leaves comments. It catches obvious issues and improvement opportunities before any human reviewer looks at it.

Trigger: GitHub PR event

Scenario 3: Weekly Content Research Automation (for Bloggers)

Automatically gather the past week's AI tool updates, generate three blog post titles and summaries, and save them to a Notion database. Wake up on Saturday to find ideas already organized.

Trigger: Scheduled every Sunday at 6 PM

Scenario 4: Auto-Feedback on Submission (for EdTech Platforms)

When a student submits an assignment, an API trigger fires a Claude agent to generate a draft feedback response. The instructor reviews and sends it — dramatically cutting personal feedback time for a class of 30.

Trigger: API (assignment submission event)

Scenario 5: Error Log Analysis and Alerts (for SaaS Services)

When error logs accumulate → Claude automatically analyzes them → sends a cause summary and resolution suggestions to Slack. An incident that happened at 3 AM has a full analysis report waiting when you arrive at work.

Trigger: GitHub event (error log update)

5. Costs, Limits, and What to Watch Out For

Current Status: Research Preview

As of April 2026, Routines is in research preview. Behavior, limits, and the API surface may change before general availability.

Plan Availability:

Plan	Routines Access
Free	Not available
Pro ($20/month)	Available
Max, Team, Enterprise	Available

What to Watch Out For:

Cloud execution means sensitive codebases should be security-reviewed before connecting
Execution time and frequency limits apply per plan
Always test in a sandbox environment before connecting to production
As a research preview, pricing and feature policies may change before general release

Closing Thoughts

Routines is not just "one more automation feature." It marks Claude Code's transition from a tool that works when you use it to an agent infrastructure that works without you.

From an EdTech perspective, this is significant. Repetitive, structured tasks — educational data analysis, feedback automation, content research — can now be handed off to AI agents that run around the clock.

Start with one small Routine. The experience of waking up to an AI-prepared news briefing already changes how you think about your workday.

Related Posts

What task would you automate first with Routines? Share your ideas in the comments!

Sources

Claude Code Routines Complete Guide — 3 Ways AI Works Even When Your Computer Is Off

Table of Contents