---
title:

Cursor Automations — from 'agent assists' to 'agent works in parallel'

date: 2026-03-05
tags: [#cursor, #automations, #ai, #agents, #development, #mcp, #slack, #github, #devops ]
draft: false
---

https://cursor.com/blog/automations

Cursor, apparently inspired by OpenClaw, has taken a swing at a paradigm shift. They released Automations — always-on cloud agents that react to events (Slack, PRs, PagerDuty, etc.), review code, write tests, triage bugs, and compile reports.

Essentially, this allows building a “development factory” on cloud agents that constantly monitor the codebase, fix issues, and handle routine tasks. This is a shift from “agent helps the developer” to “agent works in parallel with the developer”.

And you can apply this not only to code. Everything is limited only by integrations. For example, this post was written in Slack, translated into 3 languages, and delivered to various blogs via automation. It’ll be interesting to see how the formatting breaks in the process.

How it works

Agents are triggered on schedule or by events: message in Slack, new ticket in Linear/Jira, push/PR in GitHub, incident in PagerDuty, webhook.

When launched, the agent spins up a cloud sandbox (isolated execution environment), uses configured MCP servers and models, verifies its result, and maintains “memory” — learning from past runs.

Scenarios

Review and monitoring

Security review on every push to main — agent searches for vulnerabilities in the diff, ignores already discussed issues, sends serious findings to Slack.

Agentic codeowners — evaluates PR risk and either approves minor changes itself or assigns reviewers based on contribution history, logs decisions to Notion via MCP.

Incident response — when an incident occurs in PagerDuty, agent checks logs via Datadog MCP, recent code changes, and immediately sends a fix proposal and PR to Slack.

Chores

  • Weekly digest of repo changes in Slack.
  • Daily review of recent code with adding missing tests and running them before PR.
  • Triage bug reports from Slack: find duplicates, create task via Linear MCP, attempt fix, and respond in thread.

Example from Rippling

Engineer drops notes, TODOs, and links into one Slack channel. A cron agent every 2 hours collects everything together with PRs, tasks, and mentions into a unified dashboard. Plus auto-creation of tasks and summaries of discussions in Confluence.

Best automations are replicated across the team.