Alessandro Sillato

A few weeks ago, Anthropic quietly released something that I think deserves more attention than it got. It is called Routines, and it is part of Claude Code. On the surface, it looks like a scheduling feature. But if you look a little closer, it is actually a different way of thinking about what automation means.

What Routines Are, in Simple Terms

The idea is straightforward. You define a task once, connect it to your repositories and tools, and then Claude runs it automatically. On a schedule, when an API call arrives, or when something happens on GitHub. You do not need to keep your laptop open. You do not need a server. Anthropic handles the infrastructure.

Some examples of what people are already doing with this: automatic PR reviews, nightly summaries of open issues, deployment checks after a release, documentation updates triggered by code changes.

The interesting part is not the scheduling itself. Cron jobs have existed for decades. What is different here is that you are not giving Claude a list of steps to execute. You are giving it a goal. And Claude decides how to get there.

Where n8n Fits In

If you work with automation, you probably know n8n. It is a workflow tool that lets you connect services visually, without writing much code. You build a flow: this happens, then that happens, then this other thing happens. It is powerful, flexible, and you can self-host it, which many people do precisely to avoid depending on external services.

n8n is very good at connecting things. If you need to take data from one place, transform it, and send it somewhere else, n8n does that job extremely well. The logic is explicit. Every step is visible. You are in control.

The difference with Claude Routines is not that one is better than the other. It is that they represent two different philosophies.

With n8n, you define the process. With Claude Routines, you define the outcome and let the AI figure out the process.

This shift sounds small, but it is not. When you give an AI a goal instead of a script, it can handle situations that a fixed workflow cannot. If something unexpected happens mid-run, Claude does not just fail and stop. It tries to reason around the problem. That makes it more resilient, but also less predictable.

The Real Tension

Here is where I get genuinely excited, but also a little cautious.

The promise of Routines is that automation becomes more intelligent. Instead of building rigid pipelines that break when the world changes, you describe what you want and the AI adapts. That is a meaningful step forward. For anyone who has spent hours debugging a workflow because one upstream API changed its response format, this feels like relief.

But there is a dependency here that I think we should not ignore.

Claude Routines run on Anthropic’s cloud infrastructure. That is not a small detail. It means that if Anthropic’s systems go down, your automations stop. It means you are trusting a third party with the execution of processes that might be critical to how you work. It means that pricing decisions, policy changes, or service interruptions are outside your control.

This is not unique to Anthropic. Any cloud service has this problem. But with traditional automation tools like n8n, you have a real alternative: you can run it on your own server. Your automation infrastructure lives where you decide it lives.

With Claude Routines, at least for now, you cannot do that.

Is There a Local Alternative?

This is the question I keep coming back to.

The honest answer is: not really, not yet. You can run open source language models locally, and there are tools being built around them for agentic tasks. But nothing today offers the same level of integration, reliability, and ease of use that a hosted service like Claude provides. Local models are getting better quickly, but running a sophisticated AI routine on your own hardware still requires technical setup that most people are not ready for.

The gap will close. It always does. But right now, if you want the intelligence of a model like Claude handling your background workflows, you are paying for it with dependency on external infrastructure.

What This Means in Practice

If you are evaluating whether to use Claude Routines or stick with something like n8n, I think the honest answer is that it depends on what you value more.

If control and independence matter to you, n8n self-hosted is still the right choice for most workflows. You own the logic, you own the infrastructure, and you are not subject to anyone else’s uptime.

If you want automation that can handle ambiguity and reason through unexpected situations, Claude Routines is genuinely interesting. The trade-off is that you are handing some control to Anthropic.

The most realistic path, at least for now, is probably not choosing one over the other. It is using each for what it does best. n8n for structured, deterministic workflows where you need control. Claude Routines for tasks where intelligence matters more than predictability.

A Different Way of Working

What I find most interesting about this release is not the feature itself. It is what it signals.

We are moving toward a model where AI does not just answer questions or help you write things. It runs in the background, takes actions, makes decisions, and gets things done while you are asleep. That is a fundamentally different relationship with technology.

Whether that sounds exciting or unsettling probably depends on your personality. For me, it sounds exciting, with the caveat that I want the option to eventually run all of this on infrastructure I control.

The tools are not there yet. But the direction is clear.