New Free guide every week. Subscribe to never miss one. See all free guides
AI Agents

Claude just dropped routines. Here's how to put your business on autopilot.

Routines let a Claude agent run on a schedule without you sitting there. It can run your skills, use your connected tools, and deliver the result while you sleep. Here are real founder examples and an easy step-by-step to build your first one in under 15 minutes.

Most of us are still using AI like a vending machine. You walk up, type a request, wait for the answer, and walk away. The work only happens when you are sitting there to ask for it. That means the AI is only as consistent as you are, and you are running a business, so you are not consistent. Nobody is.

Routines change the shape of that completely. A routine is a saved Claude agent that runs on a schedule, on its own, on Anthropic's cloud, even with your laptop closed. You write the instructions once. You tell it when to run. From then on it shows up every morning, every Monday, or every night and does the work, then hands you the result. It can run the skills you have already built, pull from your connected tools like Slack, Google Drive, or your email platform, and even call outside services through a custom environment.

This is the difference between AI that helps you work and AI that does the work for you. It is the closest thing to hiring a quiet, reliable team member who never forgets, never gets distracted, and costs you nothing extra on your existing plan. Below are the examples worth stealing, then the exact steps to build your first routine today.

Real routines you can build this week

These are not hypotheticals. Each one pairs a schedule with a clear job and a clear deliverable, which is exactly what a routine is built for. Start with one. The first is the easiest to get running.

Every morning

Daily inbox triage

Runs at 6am, reads every email that came in overnight, sorts the urgent from the noise, drafts replies to the easy ones, and posts a short "here's your day" summary to your Slack or your inbox before you finish coffee.

Every Sunday

Weekly newsletter planner

Pulls your best-performing content and recent saves, then hands you three newsletter angles, a subject line for each, and a rough outline so you never open Monday to a blank page again.

Every weekday

Customer support cleanup

Runs your support skill against new tickets, tags them, drafts responses for the repeat questions, flags anything that needs a human, and drops a daily summary in your team channel.

Every Monday

Sales and product data report

Reads last week's numbers from your store or dashboard, calls out what moved and what stalled, and writes you a plain-English summary with one recommended action, no spreadsheet diving required.

Every Friday

Content idea scout

Scans the newsletters, creators, and free resources you follow, then delivers five fresh content ideas tied to what is actually happening in your industry that week.

Once, later

The follow-up that never slips

A one-off routine you set for two weeks out: "remind me to check in with this lead" or "open the cart-close email." It fires once at the exact time, then turns itself off.

Notice the pattern. Every good routine has a trigger (when it runs), a clear instruction (what to do), and a place the result lands (your Slack, your inbox, a doc). Hold onto that shape. You are about to build one.

The setup.

Step 01Open routines and start a new one

Go to claude.ai/code/routines and click New routine. You can also create one right from the Claude Desktop app by clicking Routines in the sidebar, then New routine, and choosing Remote so it runs in the cloud instead of on your machine.

One note before you start: routines are available on the Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled. If you do not see routines yet, that is the box to check first.

Step 02Write the instructions like you are briefing a new hire

This is the most important part of the whole thing. A routine runs alone, with nobody there to clarify, so the instructions have to be complete on their own. Tell it exactly what to do, where to get the information, and what a finished job looks like.

Vague instructions get vague work. Specific instructions get something you can actually use. Here is the shape to follow.

Paste this into the instructions box and edit the brackets

Every morning, read all emails received since yesterday in my connected inbox. Sort them into urgent, needs a reply, and can wait. For anything in "needs a reply" that is straightforward, write a draft response in my voice. Then post a summary to my [Slack channel name] in this format: top 3 things that need me today, number of emails handled, and any drafts waiting for my approval. Keep it short and skimmable.

While you are here, pick the model you want it to run on. Claude will use that same model on every run.

Step 03Connect the tools and skills it needs

A routine can reach the same connected tools you already use with Claude, like Slack, Google Drive, your email platform, or your project tracker. It can also run any skill you have built. This is what makes it powerful: you are not rebuilding your work, you are scheduling the work you already set up.

  • Under Connectors, every tool you have connected is included by default. Leave the ones this routine needs and remove the rest, so it only has access to what the job requires.
  • If your routine should run a skill, make sure that skill is available to the session, then name it directly in your instructions, for example "run my support-cleanup skill."
  • If it needs to reach a service outside the standard list, that is where a custom environment comes in. You can set network access and add the keys it needs to call an outside tool.

Step 04Set the trigger: when should it run

Under Select a trigger, pick how it starts. For most founder routines you want Schedule. Choose hourly, daily, weekdays, or weekly, and set the time in your own time zone. Claude converts it automatically, so it runs at your wall-clock time no matter where the cloud lives.

You are not limited to one. A single routine can combine triggers, and the two you will reach for most are these.

  • Schedule, for anything recurring: the morning triage, the Sunday planner, the Monday report.
  • One-off, for a single future moment: set it in plain language like "in two weeks, open the cart-close email." It fires once, then disables itself.

There are also API and GitHub triggers for more technical setups, but you do not need those to get real value on day one.

Step 05Create it, then run it once on demand

Click Create. The routine now lives in your list and will run the next time its trigger matches. Do not wait until tomorrow morning to find out if it works.

Open the routine and click Run now. This starts a run immediately so you can watch what Claude actually did, read the result, and confirm it landed where you wanted. Treat this first run as a test drive.

Step 06Read the first run and tighten the instructions

Open the run and read the full transcript, not just the green status light. A green status only means the session started and finished without crashing. It does not mean the job was done the way you wanted. The transcript is where you see what it really did.

If something was off, the fix is almost always in the instructions. Add the detail it was missing, name the exact format you want, or point it at the right source. Edit the routine, save, and run it once more. Two or three rounds of this and you will have a routine you trust to run without you.

That's it.

You just built an AI agent that runs on a schedule, uses your tools, and delivers the result while you are doing something else. Start with one routine, the morning triage or the Sunday planner, and let it earn your trust. Then add the next. Each one you build is time you stop spending on the same task every single week.

The founders pulling away right now are not the ones asking AI better questions. They are the ones who stopped asking and started scheduling. That is exactly the kind of system we build inside Her AI Systems™.

Common questions.

What plans include Claude routines?

Routines are available on the Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled. If you do not see routines yet, that is the setting to check first.

Do Claude routines run when my computer is off?

Yes. Routines run on Anthropic's managed cloud infrastructure, so they keep working even when your laptop is closed.

Can a routine use my skills and connected tools?

Yes. A routine can run the skills you have built and use your connected tools, such as Slack, Google Drive, your email platform, or your project tracker. All your connected tools are included by default, and you remove the ones a routine does not need.

How many routines can I run per day?

Pro users can run up to 5 routines per day, Max users up to 15, and Team and Enterprise users up to 25. One-off runs do not count against this daily cap.

What is the difference between a recurring routine and a one-off task?

A recurring routine runs on a repeating schedule like daily or weekly. A one-off fires once at a specific future time you set in plain language, then disables itself automatically.

Free Download

Save this guide as a PDF.

Your future self will thank you for saving this.

Want more?

Get every new free guide in your inbox.

One new free guide. Every Tuesday. No spam, no fluff, just the systems and prompts I'm building for clients, simplified for you.

Subscribe

The FREE Her AI Systems Summit drops Summer 2026. Be the first to know when dates go live. Join the Waitlist →