Claude Routines: the AI tool that works while you sleep
A non-developer's guide to the Anthropic launch most people will miss
Claude just got the ability to work without you sitting there. Anthropic released Routines today, and for anyone running a business alone, it’s the most practically useful AI launch of 2026 so far. Below: what it does, how to use it if you’re not a developer, four routines worth building this week, and where it still breaks.
Who this is actually for?
If you are not a developer, this is written for you.
Most of the internet will cover Routines as a developer release. On the surface, it is one. But the real story is what the rest of us can finally do with it.
If you run a newsletter, a coaching business, an online store, a consulting practice, a one-person agency, or anything where the operational grind eats more of your week than the creative work, keep reading.
If you are a developer, some of this will feel obvious. Skip to the “Four routines” section, or send this to a non-technical friend who keeps asking what the AI hype is actually useful for.
What is a Claude Routine?
A routine is a Claude task that runs on Anthropic’s servers while your laptop is closed. You set it up once. You tell Claude what to do, which apps it can use, when to fire, and where the results should land. It runs. You come back to results.
That is the whole product.
Setup note: Routines lives inside Claude Code, not the regular claude.ai chat interface. If you have only been using Claude in the web or desktop chat, this feature is not there for you yet. You need Claude Code (either in your browser or installed on your computer) to configure and run one.
The setup looks technical at first. It is not as bad as it looks. And it changes how your week works in a way I will explain in a minute.
How a routine knows when to run
Three options. You pick whichever fits the job.
Schedule it. You tell Claude when to run. Every weekday at 9am. Every Sunday at 8pm. Every first of the month. Whatever cadence the work needs. Under the hood this uses a format called cron, which looks intimidating if you have never seen it. In practice you just set a time and Claude handles the rest.
Trigger it from another app. Every routine gets its own dedicated web address. Any tool in your stack that can send a signal can hit that address and wake the routine up. This is how you connect Routines to Stripe, Calendly, Typeform, Shopify, your email platform, or anything else your business runs on.
GitHub events (for developers only). If you use GitHub for code, Routines can fire directly on GitHub activity without any middleman. If that sentence means nothing to you, skip this option. You are not missing anything relevant to your business.
Why this matters for anyone running a business solo
If you run a business by yourself, most of your week is not the fun part. It is the operational grind: admin, follow-ups, research, reformatting the same information for the fifth time, moving data between tools. Nobody brags about any of that on LinkedIn. But it eats your week.
The old loop with AI tools: they only worked when you were there. You opened the app. You prompted it. You waited. You reviewed the output. The moment you stepped away, the productivity stopped.
Routines end that loop.
Example. Every night at 2am, a routine pulls your saved content ideas from Notion, researches the top one, drafts an outline in your voice, and drops it into a “ready to review” database. You show up the next morning. The thinking is done. You edit and ship.
Anyone who has opened their laptop in the morning and wished half the work was already there knows what that shift feels like.
How non-technical people connect this to their tools
Now the honest part.
Anthropic has built only one native event trigger so far, and it is GitHub. Your Stripe sales, Typeform submissions, Calendly bookings, and Shopify orders cannot fire a routine directly yet.
But they can fire a routine indirectly. The flow looks like this:
An event happens in Stripe (or Calendly, Typeform, Shopify, whatever you use).
You use Zapier, Make, n8n to catch that event.
Zapier sends a signal to your routine’s dedicated address.
Claude wakes up and does the actual work.
So Zapier is not dead yet. What Zapier does well (catching events from one app and forwarding them somewhere else) still matters. What Zapier does badly (the 8-step chain of filters, formatters, AI calls, and manual cleanup) gets replaced by a single Claude routine that actually thinks.
The logic moves to Claude. The plumbing stays where it is. For now.
Four routines worth building first
Quick note before the list: for Claude to reach into any of your tools, you need something called a Connector set up in your Claude Code environment. A Connector is basically permission for Claude to talk to one of your apps (Notion, Gmail, Google Drive, Slack, and more coming). The Connector list is growing. You cannot ask a routine to check a tool you have not wired up yet.
With that said:
1. New lead research. Someone fills out your contact form. Zapier catches the submission and pings your routine. Claude researches the lead using public info (their LinkedIn, any prior interactions you have had with them). Drafts a custom reply in your voice. Queues it in your drafts folder. You review and send. Total time: 90 seconds instead of 20 minutes.
2. Morning intelligence brief. Scheduled at 6am every weekday. Claude pulls yesterday’s analytics, overnight replies, news on the 5 competitors you track, and any open loops from last week. Summarizes everything into a 2-minute read. Drops it in your Notion dashboard. You drink coffee and get oriented before the day starts.
3. Idea processor. Scheduled overnight. Claude reads through the notes, voice memos, saved articles, and screenshots you flagged during the day. Organizes them by theme. Builds the best ones into rough drafts you can develop the next morning. No more ideas dying in Notion graveyards.
4. Weekly business review. Runs every Monday at 7am. Claude pulls revenue data, subscriber numbers, pipeline stage changes, and invoicing status from across your Connectors. Writes a one-page state-of-the-business brief. You walk into Monday already oriented.
Each of these used to require a fragile chain of Zapier filters, a stored prompt somewhere else, a hand-rolled AI call in step 7, and you at the keyboard to clean up the output. Now the brain of the whole operation is one Claude routine.
Why this beats what OpenClaw has shipped
Quick context on why this matters in the wider AI space.
OpenClaw has been making noise about agents, GPTs, Assistant API runs, and custom tools for about a year now. Most of it has landed as chat-surface demos. Cute. Not very useful for actual work.
Anthropic just shipped real infrastructure: AI that runs on their servers, on a schedule or triggered by signals from your other tools, with authentication and logging handled for you.
That is the difference between a product demo and production plumbing. Which is what Anthropic has been quietly accumulating while everyone else was chasing viral screenshots.
Routines is the least-hyped big launch I have seen from Anthropic this year and probably the most practically important. Which tells you exactly where the company’s attention actually sits.
Where this will break
Honest limits before you wire this up to your whole life.
Your routines are only as good as the context you give them. A routine that drafts content in “your voice” from a generic prompt will produce the same slop everyone else is producing. The infrastructure is shipped. The voice problem is still yours to solve.
This is a research preview. Things will break.
Usage limits are real. Claude Pro gets 5 runs per day. Max gets 15. Team gets 25. You are not running unlimited background automation yet. Pick your routines carefully.
Native event triggers are GitHub-only for now. Everything else needs Zapier or Make sitting in between to relay the signal.
Don’t run anything irreversible through a routine yet. No auto-sending client emails. No firing invoices without a human review step. Draft and queue. Don’t send and pray.
How to build your first routine
Start with one. Don’t try to automate your whole operation on day one. You will build something brittle, produce garbage for a week, rage-delete it, and walk away angry.
Pick the single task that makes you tired before you start. Lead research. Morning catch-up. Weekly numbers pull. Idea processing. Whichever one drains you the most.
Build one routine that does 60% of it while you sleep. Wake up to work that is already started. See how it feels.
If it feels good, build another next week.
If Claude doesn’t have your voice or context yet, that is a separate problem. It is also the one I am writing about next.
The chatbot era is over. The operator era just started.
What is the single task in your business you would fire overnight and wake up to completed? Hit reply and tell me. I read every one.
FREE SUBSCRIBERS: 2+ issues per week. Frameworks, prompts, operator thinking, systems thinking. The “what” and the “why.” Enough to change how you approach AI and work forever.
PAID SUBSCRIBERS ($15/month): Everything above, plus the implementation. The “HOW.”
Copy-paste prompts I use daily
Complete operator systems with step-by-step builds
Access to all previous issues with templates, frameworks, and resources
Direct access (I read every reply)
One system that saves you 2 hours per week = 100+ hours per year. You get multiple systems. Every week. Founding rate locked forever.
Most individuals won't notice this change until it becomes the norm. The functionality isn't what makes it valuable; it's that solitary operators can finally get rid of the boring, context-heavy job that usually takes up their whole week. When AI remains working after you close your laptop, the limit on what one person can do changes altogether. The demo was the time of chatbots. This is the structure.