A discussion on Hacker News has soared in recent hours around Anthropic's new «Routines» feature in Claude Code. These are early signals from a tech community that is usually quite skeptical, so it's worth noting that the mood here is unusually positive.

So what exactly are Routines? In short: you define a reusable workflow once — in the project's CLAUDE.md file or via the configuration system — and then Claude Code can execute it on command, or completely automatically. What makes this interesting is that the jobs run on Anthropic's own servers. No local process, no open terminal. You can initiate a nightly job that scans docs drift or cleans up the backlog, and Claude just... does it.

Anthropic claims that the majority of its own code is now written by Claude Code.

There are three types of Routines: Scheduled (run this every night at 02:00), API-triggered (POST to an endpoint and start a session), and Webhook-based (connect directly to GitHub and let Claude react to pull requests or CI errors automatically). The latter is what really gets people talking — this is starting to resemble a full-fledged agentic development environment, not just a fancy autocomplete tool.

Compared to OpenAI's Code Interpreter, the difference is quite clear: Code Interpreter is strong for ad-hoc data analysis in a sandbox, but has no internet access, no persistence between sessions, and no support for external API integration. Claude Code Routines are built for something else — long-running, autonomous development tasks deeply integrated into an actual code repository.

Limitations exist, of course. Pro users only get 5 Routines per day, Max users 15, and Team/Enterprise up to 25. It's not unlimited power, and Claude Code still requires explicit permission before it actually modifies files or runs commands. Humans still decide what ships.

But the direction is clear: Anthropic is no longer just building a coding assistant. They are building an autonomous developer that lives in your infrastructure.

Important disclaimer: This is based on community discussions and documentation from Anthropic, not independent verification. As always with early signals — pay attention, but wait for more documentation before basing critical decisions on this.