Key insights:
Coding agents shouldn't wait for you to press enter.
Right now, most developers treat Claude Code as a powerful coding tool. You type a prompt, hit enter, and wait for the result. But what if Claude could act like a real teammate instead? A teammate notices when something breaks and does something about it.
That's the shift Anthropic is pushing with a new feature called routines. It moves Claude Code from a reactive tool into a proactive teammate that reads your repo, runs on a schedule or event, and opens a PR before you've even opened your laptop.
A tool sits idle until you use it. A teammate works alongside you, anticipating problems and acting on them.
Think about your own workflow. How often do you repeat the same investigation, the same review, the same documentation update? Each of those is a candidate for automation.
With proactive agents, you stop being the trigger. The agent watches your repo, your deploys, or your issue tracker and kicks itself off.
Building a proactive agent today means building a lot of infrastructure. You need hosting. You need data persistence. You need authentication. You need a way to trigger sessions reliably.
Running Claude Code on a cron job sounds simple. In practice, it means managing a server, handling failures, and maintaining boilerplate code that has nothing to do with your actual problem.
Most developers want to focus on what the agent does, not on where it lives or how it starts.
Routines solve three core challenges with proactive agents:
You define the prompt, the repos, the connectors, and the trigger. Claude Code handles the rest.
Routines are a new feature inside Claude Code that let you launch remote Claude Code sessions automatically. Every routine runs on managed infrastructure, so nothing depends on your local machine.
You start with a single command inside Claude Code: /schedule.
From there, you describe what you want. For example, you might type: "Once a week, review all the new changes merged to main against our documentation repo and create a PR to update docs if you see any gaps."
Claude follows up with questions. What time do you want this to run? Should it notify you on Slack? Once you answer, Claude creates the routine for you.
You can then view and edit the routine on claude.ai under the code section.
Routines support two main types of triggers:
GitHub events are supported natively. You can trigger on a new issue, a merged PR, a release cut, or a labelled PR.
For custom events, you post to a webhook with your own event payload, and the routine kicks off with that context.
Context is the ceiling of what your agent can do. Without the right information and tools, Claude can't be successful.
You can connect one or more code repositories so Claude can read the source and open PRs. You can also attach connectors like Google Drive for documents, Slack for notifications, or the GitHub MCP for issue management.
Think carefully about what Claude needs:
At Anthropic, weekly PRs for Claude Code have gone up 200% since the start of the year. That's great for engineers and great for users who get new features quickly. It's not so great for the single engineer responsible for keeping documentation in sync.
The first routine is simple. Once a week, Claude reviews all changes merged to the Claude Code source repo and compares them against the documentation repo.
If Claude finds gaps, it opens a PR in the docs repo with the updates. It pings the maintainer on Slack so they know to review.
The trigger is a weekly schedule. The context includes the source repo, the docs repo, GitHub, and Slack. The output is a PR ready for human review.
The second routine reacts to new GitHub issues opened on the docs repo.
The instructions tell Claude to investigate the issue, decide if it's a real documentation gap, and if so, open a PR and notify the team on Slack.
The trigger is a GitHub event for issue opens. The context is the same two repos plus Slack and the GitHub MCP. The session begins with the issue body passed in as context, so Claude knows exactly what to look at.
This pattern works for any repo where users file gaps as issues. Claude becomes the first responder.
Routines aren't fire and forget. Every routine is a Claude Code session you can open, watch, and steer in real time.
You can jump into the session on the web, read Claude's analysis, and push it in a different direction. You can stop a session early if you already handled the issue. You can resume a past routine and continue the conversation.
This matters because sometimes you want a human in the loop and sometimes you don't. Routines give you both.
Routines work for far more than documentation. Any repeatable workflow that involves a trigger, some context, and a decision is a candidate.
Imagine you just deployed a change to a service and you want to make sure it's healthy.
The trigger could be a webhook from your CD pipeline that fires after every deploy. The context would include the source code for the service, monitoring tools like Datadog or Grafana, and a notification channel like Slack or Twilio.
Claude investigates the deploy, reads the metrics, and gives you a go or no-go decision. Over time, as you trust it more, you can let Claude roll back the change itself if the data warrants it.
Product managers spend hours sifting through GitHub issues and Slack threads. A weekly routine can do that work for you.
Trigger it on a Monday morning schedule. Give it access to GitHub, Slack, and your issue tracker. Ask Claude to read every new issue, group them by theme, prioritise the most urgent, and post a summary in a channel.
You can take this further by having Claude open draft PRs for the highest priority bugs, so engineers walk in to ready work.
On-call work is full of repeatable investigation patterns. When an alert fires, you check logs, look at recent deploys, and form a hypothesis.
A routine can do the first pass. Trigger it on a PagerDuty webhook. Give it access to your logs, your repo, and your monitoring tools. Have it post a triage summary in the incident channel within minutes.
You still make the final call. But you skip the boring first ten minutes of every incident.
Routines are a sign of where coding work is heading. The developers who win in the next few years aren't the ones racing AI on raw output. They're the ones designing the agents, the triggers, and the workflows that AI runs inside.
Designing a good routine isn't just about prompts. It's about thinking like an automation engineer. You need to understand triggers, context, tools, error handling, and how to keep a human in the loop when it matters.
That's exactly the skill set taught inside the Complete RPA Bootcamp. You go from beginner to pro across Robotic Process Automation, Agentic Automation, Coded Automation, and Computer-Use Agents.
Instead of letting AI and automation replace you, you become the one building the AI and automation. It's a future-proof career path designed for this exact moment.
Inside the bootcamp you build real, working automations end to end. You learn how to:
These are the same patterns Anthropic uses internally with routines, applied across a much wider range of business and technical problems.
Yes. Tools like Claude Code routines are making it possible for one developer to deliver the work of an entire team. Companies are noticing, and they need people who can design and ship these systems.
You can either watch this shift happen or be the person making it happen. The Complete RPA Bootcamp gives you the structured path to do that, without piecing together random tutorials.
To see routines in action and watch a full proactive agent workflow built end to end, watch the embedded video below from the Claude YouTube channel. It's the clearest walkthrough you'll find on turning Claude Code from a tool into a true teammate.