Practical ways to use Claude Routines at work - a Guide for Product Teams

🧠 Backlog triaging, API deprecation watchdog, user feedback aggregation and more.

🔒 The Knowledge Series breaks down emerging AI technologies with practical playbooks designed specifically for product teams. Get 100+ guides and practical tutorials covering everything from Claude Code and MCP to agentic workflows, vibe coding, and more.

---

At Anthropic’s recent Claude Code events in London and San Francisco, alongside the talks announcing new features and capabilities, many of the speakers took their time to re-engage with features that already exist.

Given the sheer velocity of new features shipped by Anthropic and other AI companies over the past few months, these refresher talks were a welcome addition to the lineup where Anthropic's employees revisited some of the most powerful tools available across the Anthropic product suite.

And one of the products that was mentioned and demoed by multiple speakers was Claude Routines.

Claude Routines was originally released earlier this year as part of a full redesign of the Claude Desktop app but since then it has often found itself drowned out by other flashier new releases. But, getting it set up as an always-on assistant to work on regular routines on your behalf can be a powerful addition to your everyday workflow.

In this Knowledge Series, we’ll step through the essentials of Claude Routines and how it works, together with 5 practical routines you can get started with, including:

1. Weekly backlog triage

2. Changelog to release notes pipeline

3. API deprecation watchdog

4. Competitor changelog tracker

5. User feedback aggregator

---

The Knowledge Series

Claude Routines explained; what they are and why they’re valuable

Routines is a feature inside Claude Code that lets you run automated, proactive Claude Code sessions without building any infrastructure yourself. In a sense, they let you put Claude Code on autopilot - with some caveats.

Instead of waiting for you to type a prompt and press enter, Claude can kick off work on its own - triggered by a schedule or an external event.

Traditionally, running Claude Code in an automated way with cron jobs or a webhooks would mean you’d have to manage everything yourself, and if your laptop closes, the session dies. Routines moves all of that onto Claude Code’s managed infrastructure so none of it depends on your local machine - unless you want it to run locally, instead.

How to set up a routine

There are two ways to create a routine. From the terminal, you can type /schedule inside Claude Code and describe what you want done in plain language. Claude asks a few follow-up questions - when should this run, what tools does it need, where should it send notifications - and then creates the routine for you.

Alternatively, in the Claude Code desktop app, navigate to Routines in the left sidebar and open the “New routine” form.

You fill in a name, write the instructions directly into a text box, pick a trigger from a dropdown, select which repositories to connect, and attach any connectors the routine needs - such as Google Drive, Notion, or Slack. The form gives you a bit more visibility upfront, since you can see and edit all the configuration in one place before hitting Create.

For this tutorial, we’ll stick to the Claude Code desktop app to keep things simple.

A routine needs to have the following options set:

• Name - a label for the routine

• Instructions - the prompt Claude follows during each run

• Repository - this is one or more codebases Claude has access to. For Routines to work remotely they need to have access to the relevant code base/s. More on that below.

• Trigger - the event or schedule that kicks off the routine

• Connectors - external integrations available to Claude during each run

• Behavior - a tab suggesting there are settings around how Claude operates (not fully visible)

• Permissions - a tab suggesting there are controls around what Claude is allowed to do (not fully visible)

Local vs Remote: How repositories work in the context of Routines

When setting up your routine, you’ll be given the choice between running it locally or remotely. Since we don’t want these routines to dependent on our machine running, we’re going to choose remote.

Claude Routines is designed to work with code repositories and so to get the most out of it, you’d need to be able to give Claude access to one or more repositories.

Example of creating a new repository in GitHub called “claude-routines-test”.

In our examples, we’ll assume that you have access to your a code repository and a place where you can save skills to be used in Routines.

How triggers work

There are two types of triggers when working with Routines: time based and event based triggers.

Time-based triggers run on a schedule. So, for example, every Monday at 10am, once a week, or whatever cadence makes sense for the task.

Event-based triggers fire in response to something happening. Within this category there are two options:

• Native GitHub events, such as a pull request being opened or closed, an issue being created, or a release being cut.

• Custom events from your product to an API, where you post to a endpoint that routines expose. You can include whatever context you want in the API request, and Claude will have access to it during the session.

How connectors work in the context of Routines

Connectors also play a critical role in Routines. Connectors are integrations you attach to a routine that give Claude access to external services during each run.

When you set up a routine, you choose which connectors to include - things like GitHub, Slack, Google Drive, or Notion. These define what Claude can actually reach out to while the session is running. For example, if you want Claude to open a PR when it finds a documentation gap, it needs the GitHub connector. If you want it to ping you when it’s done, it needs Slack.

One thing that’s worth noting is that Claude can use all tools from attached connectors - including write operations - without asking for permission during each run. That’s by design, since these are meant to be autonomous sessions that don’t wait for human approval. But it means you should be deliberate about what you attach. If you give Claude a connector, assume it may use it.

This is different from a regular Claude Code session where you’re present and can approve or reject actions as they happen. In a routine, Claude is operating on its own, so the connector list acts as the boundary of what it’s allowed to touch.

5 Practical ways to use Claude Routines at work

Now that we’ve covered some of the essentials worth knowing, let’s take a closer look at some of the practical ways you can actually use Claude Routines at work.

For this, we’ll explore 5 different use cases including a weekly backlog triage, changelog to release notes pipeline, API deprecation watchdog, competitor changelog tracker and user feedback aggregator.

1. Weekly backlog triage