How Claude and Cursor’s New Automations Eliminate Midnight Sentry Checks

In the early hours, Claude announced Claude Code Routines while Cursor posted about Cursor Automations, both enabling cloud‑based automatic responses to Sentry alerts so developers no longer need a running computer to monitor errors.

Cursor’s solution is a marketplace automation template called "Investigate and fix Sentry errors". Users select the Sentry‑new‑issue trigger, provide instructions such as "investigate my SQLite service’s Sentry error and implement a high‑confidence fix", save, and the automation runs.

The Agent follows a fixed five‑step workflow:

Fetch full error details from Sentry, including stack trace, exception type, affected endpoint, and frequency.

Group recurring symptoms into root‑cause clusters.

Correlate stack frames, request context, timestamps, and recent code changes to locate the change point.

Form a root‑cause hypothesis and validate it with code evidence.

If confident, open a PR to fix the issue; otherwise, emit an investigation report and a follow‑up plan.

The fix result is posted to Slack with error tags, root‑cause description, fix summary, verification steps, and residual risk, offering a ready‑to‑use "plug‑and‑play" experience that requires no prompt engineering.

Claude Code Routines take a more general infrastructure approach. A Routine consists of three parts: a user‑written prompt, one or more GitHub repositories, and connected MCP connectors such as Slack, Linear, or GitHub. It supports three trigger types that can be freely combined.

Scheduled trigger : runs hourly, daily, on weekdays, or weekly with automatic timezone conversion. A typical use case is pulling high‑priority bugs from Linear at 2 am, attempting a fix, and opening a draft PR so it is ready when developers start work.

API trigger : each Routine exposes an HTTP endpoint and bearer token; a POST request with optional context (e.g., an alert payload) triggers a run, allowing integration with monitoring systems, deployment scripts, or internal tools.

GitHub event trigger : listens to dozens of events such as PR, push, issue, release, or Actions workflow, with optional filters (e.g., only when a PR has a specific label or originates from a fork).

Routines execute on Anthropic’s cloud servers, independent of the developer’s local environment. After configuring at claude.ai/code/routines, the computer can be shut off while tasks continue to run.

Claude’s Sentry‑alert example mirrors Cursor’s: the monitoring tool calls the Routine’s API with the alert, Claude analyses the stack, identifies the recent commit that introduced the bug, and opens a fix PR.

Shortly after releasing Claude Routines, Claude announced a major desktop client redesign, adding multi‑session parallel support and a unified sidebar for managing all sessions, enabling developers to run multiple manual Claude conversations while cloud‑based Routines operate in the background.

The simultaneous announcements illustrate a broader shift: AI coding assistants are moving from IDE‑centric code‑completion tools toward "event‑driven autonomous agents" that can monitor environments, act proactively, and close the loop without manual invocation.