Claude Code Routines Are Hugely Underrated. Here’s How to Get Ahead of 99% of Users
7 recurring tasks worth scheduling today, with necessary tools, inputs, outputs, and starter prompts for each.
Claude Code Routines let you schedule recurring AI tasks without cron jobs, scripts, or an OpenClaw setup. This guide covers the specific use cases worth scheduling: research digests, SEO checks, content planning, inbox triage, and lightweight reports. With the exact inputs and outputs for each. It also covers when Routines are the wrong tool and what to reach for instead.
When the Claude Code redesign dropped in April, the biggest draw wasn’t the new UI.
It was Routines: a scheduling layer that promises to run your AI work autonomously. Lighter than OpenClaw. No server required. Built right into Claude.
The biggest question I got from that update was:
Is OpenClaw still worth exploring?
I get the anxiety. You bought the Mac Mini, did the install, got things running. And now Claude has a scheduling layer built in, and Hermes is chasing from the other side.
So I tested it. Migrated what I could to Routines, kept what belongs in OpenClaw, figured out where the line actually is.
And that line is more interesting than I expected.
OpenClaw was the first tool that made a lot of builders feel what autonomous AI work could become: an agent running workflows in the background, not just answering you in a chat box. But it also came with setup, hardware, architecture, and one more system to maintain.
Claude Code is now closing part of that gap.
Not all of it. But for recurring work like research digests, content planning, SEO checks, and project summaries, Routines (together with every Claude scheduling features) get you surprisingly close with less setup and inside a subscription you may already be paying for.
That is the real change. You can now get a meaningful version of autonomous AI work without building the whole automation stack first.
In this guide, you will see:
Where Claude Code Routines Fit — the quick version of why they matter
7 recurring tasks worth scheduling right now — from a 2-minute demo to a full research-to-planning stack
The first routine I’d build — the one that shows the return fastest
Hi, I’m Jenny 👋
I build AI systems and tools, then share how I did it. I run the Practical AI Builder program — for people who already use AI and want to build real things with it. Check it out if that sounds like you.
If you’re new to Build to Launch, welcome! Here’s what you might enjoy:
Where Claude Code Routines Fit
If you already read the Claude Code redesign breakdown, you can skip ahead. This is the quick version for everyone else.
Routines are scheduled tasks inside the Claude Code tab of the desktop app.
You give Claude a project, an instruction, a schedule, and an output location. Then it runs the task without you opening a live chat.
The important part is that this is code-first automation. Claude can read files, use your local context, run logic, and write the result back into your repo.
That puts Routines at Level 3 in my 4 Levels of AI Automation framework: above one-shot Claude sessions, below full OpenClaw pipelines.
Use them when the job is:
repeatable
file-based
scoped enough to review after the fact
The cleanest shape is simple:
Read these inputs on this schedule. Produce this output. Tell me when it is done.
If a failed run leaves you with a missing report or stale planning file, a Routine is fine.
If a failed run sends the wrong email, deletes data, publishes publicly, charges someone, or blocks another workflow, use something heavier.
That is the whole point: Routines give you useful autonomous work without making you build the whole automation stack first.
For this guide, I am focusing on local Routines. They run on your Mac, which means your machine needs to be on, but they can use the files, repos, and MCPs you already have connected.
7 Claude Code Routines Worth Scheduling
What follows is a progression from the simplest possible Routine to more complex setups.
The first Routine is deliberately tiny. It is there to help you understand the mechanics: what it feels like to set something up once, hit Run Now, and see Claude do something without you prompting it live.
After that, each Routine teaches a reusable pattern. Once you understand how one Routine works, you can point that same shape at a different topic, tool, data source, or workflow. You can also expand on top of it.
That is the change I want to deliver in this piece: you are not just copying my automations. You are learning the building blocks for your own scheduled AI workflows.
Each Routine follows the same structure:
- what it does,
- what tools it needs,
- what goes in,
- what comes back,
- why it is useful, and
- a starter prompt you can copy and adapt.
Before scheduling any of them, run the Routine manually once and check three things:
Did it read the right inputs?
Did it write the output where expected?
Did it leave a clear completion signal?
Do not schedule anything until the manual run produces an output you would actually trust.
Routine 0: The First Routine (Demo)
This one does nothing impressive. That is the point.
What it does: Claude sends a Slack message on a schedule. That is it.
Tools needed: Claude Code. A Slack MCP connected to Claude
What goes in: A single instruction: “Send a message to #my-channel saying: Weekly routine fired at [time]. Date: [date].”
What comes back: A Slack message.
Schedule it for Monday morning. Hit “Run Now” once to confirm it fires. Then watch it run on its own next week.
Why start here: Most people never get their first Routine running because they try to build something useful before they understand how the mechanism works. This one removes that trap. Once you see a scheduled task fire on its own, even a trivial one, the rest becomes obvious.
Starter prompt:
Send a message to #[your-channel] in Slack:
"Weekly routine fired. Date: [today's date]."
The following routines are where this starts doing real work.
Each one shows a different type of data source or output:
One pulls research from the web.
One reads your inbox.
One mines your own Google search performance.
One repurposes content while you sleep.
One fixes your build docs.
One auto-adjusts posting strategy.
Different inputs, different outputs, same transferable pattern underneath each one.
Each routine tells you exactly what tools you need, what goes in, and what comes back.
Routine 1: Weekly Research Digest
Pattern: Data In — external sources
What it does: Claude scans a list of topics you care about, filters out what you have already covered, and writes a weekly brief on what is new and worth acting on.