Cursor Automations launched on March 5, 2026, and your CI/CD pipeline should be nervous. The feature turns AI coding agents into always-on background workers triggered by Slack messages, GitHub PR merges, PagerDuty incidents, or any webhook — running in cloud sandboxes without requiring a developer to open the IDE. Anysphere, the company behind Cursor, already holds roughly 25% of generative AI clients according to Ramp data cited by TechCrunch, and reportedly surpassed $2B in annualized revenue — doubling in three months. Cursor Automations is the product that explains how that happened.
This isn’t agents running in the background while you watch. It’s event-driven workflow automation with native MCP orchestration, made possible by a billing change last June that turned every triggered agent into automatic revenue.
What Cursor Automations Actually Does
The core mechanic: real-world events fire triggers, and each trigger spins up an AI agent in an isolated cloud sandbox. Supported triggers include Slack messages, Linear issue creation, GitHub PR merges, PagerDuty incidents, scheduled timers, and custom webhooks. Each agent follows user-defined instructions, uses configured MCP tools, verifies its own output, and retains memory from past runs.
Cursor’s official announcement showcases five automation types that reveal the product’s ambition. Security review runs on every push to main, auditing diffs for vulnerabilities and posting findings to Slack. Agentic Codeowners classifies PR risk by blast radius, complexity, and infrastructure impact — auto-approving low-risk changes and routing the rest to humans.
The heavier-lift automations are more telling. Incident response fires on PagerDuty alerts, queries Datadog logs via MCP, messages on-call engineers, and proposes fixes via PR. Daily test coverage reviews merged code and identifies gaps. Weekly digests summarize repository activity to Slack — all without a single human prompt.
Cursor says it already runs hundreds of these automations per hour internally. As Jonas Nelle, engineering chief for asynchronous agents, told Dataconomy: “They’re called in at the right points in this conveyor belt.” That metaphor matters. This isn’t an assistant waiting for a prompt. It’s a system that activates at predefined workflow points — closer to GitHub Actions with reasoning than to a chatbot with a code editor.
The Billing Model Behind $2B in Revenue
Cursor’s June 2025 switch from fixed “fast request” allotments to usage-based credit pools looked like a billing cleanup at the time. In retrospect, it was infrastructure for Automations. Always-on agents that fire 24/7 in response to events — whether or not a developer is at their desk — burn credits continuously from the pool. Every execution is metered revenue.
The revenue trajectory tells the story: $100M ARR in January 2025, $500M by June 2025, past $1B after the $2.3 billion Series D in November 2025, and now reportedly above $2B in annualized revenue as of February 2026 per Bloomberg figures reported by TechCrunch. The doubling period from $1B to $2B coincides exactly with internal Automations adoption. For context: Slack took seven years to reach $1B ARR, Zoom took four. Cursor went from $100M to $2B in roughly 13 months.
The consumption flywheel is baked into the pricing page. Pro ($20/month) includes cloud agents required for Automations. Pro+ ($60/month, 3x usage) is marked “Recommended” — the natural upgrade path when automated agents start burning through credits overnight. Ultra ($200/month, 20x usage) sits at the top. More automations trigger more credit usage, which creates upgrade pressure, which funds more compute for better agents. It’s elegant and relentless.
MCP Is the Engine Competitors Can’t Copy Overnight
Every major AI coding tool now offers some form of background agents. GitHub Copilot’s coding agent runs asynchronously via GitHub Actions — you assign a task to Copilot through a GitHub issue and it spins up a cloud workspace, makes changes on a branch, and opens a PR. OpenAI’s Codex manages parallel agents from its desktop app. Claude Code supports scheduled local tasks with the /loop command. None of them offers event-driven triggers with native MCP orchestration as a first-class product feature.
MCP — the Model Context Protocol — is the connective tissue that makes Automations more than a cron job with an LLM attached. Confirmed integrations include Datadog for log investigation, Linear for issue creation and triage, Notion for database logging, and Runlayer for infrastructure automation. Agents can access up to 40 MCP tools per session. If you want to understand the broader protocol ecosystem, we covered the MCP server ecosystem in depth.
The incident response use case makes the architectural difference concrete. PagerDuty fires an alert. An agent spins up, queries Datadog MCP for logs and recent code changes, messages on-call engineers in Slack, and proposes a fix via pull request. No human typed a single prompt. As Help Net Security detailed, the Agentic Codeowners system evaluates PRs across three dimensions — blast radius, complexity, and infrastructure impact — to decide what gets auto-approved and what needs human eyes.
Josh Ma, engineering lead at Cursor, put the compute philosophy plainly in TechCrunch’s exclusive: “This idea of thinking harder, spending more tokens to find harder issues, has been really valuable.” Translation: the agents aren’t just pattern-matching. They’re burning credits to reason deeply — and Cursor is billing for every token.
The Tasks Cursor Is Automating Are the Ones That Train Junior Engineers
Look at the five named automation types again: security review, PR risk classification, incident first-response, test gap coverage, documentation digests. These aren’t random workflow annoyances. They are canonical junior engineer tasks — the exact work that builds codebase knowledge, teaches failure modes, and develops engineering instincts.
Reviewing PRs teaches you how a codebase breathes. Triaging bugs teaches you how it breaks. Incident response builds instincts under pressure. When those reps go to an automated agent instead, the productivity gain is real and immediate — but the second-order effect is a narrower on-ramp for junior engineers three to five years out. We’ve explored this tension before in our analysis of AI coding assistants and developer skill development.
Tim Fall from Rippling confirmed the efficiency thesis in Cursor’s announcement: “Automations offload repetitive work — task consolidation, documentation updates, Slack responses — freeing focus on meaningful work.” Tal Peretz from Runlayer said his team “moves faster than groups five times their size.” Fewer people doing more, not more people doing the same. That’s not a cautionary tale — it’s a structural shift worth naming. Companies capturing short-term efficiency gains should be asking where their next senior engineers will come from.
Four Launches in Ten Days Signals a Land Grab
Automations didn’t arrive in isolation. According to Cursor’s changelog, the company shipped four major features in ten days: computer use for cloud agents on February 24, MCP Apps and Team Marketplaces in v2.6 on March 3, JetBrains IDE integration on March 4, and Automations on March 5. Each addresses a different strategic vector — platform expansion, developer experience, agent capability, and workflow automation — from a team of roughly 300 people whose in-house models reportedly generate more code than nearly any other LLM globally.
For readers evaluating whether to adopt Cursor now, we compared how Cursor stacks up against Claude Code and Windsurf before Automations existed. That calculus just shifted significantly. Automations is available now at cursor.com/automations, with templates at cursor.com/marketplace. Cloud agents require the Pro tier at minimum — $20/month.
But here’s the question nobody in the launch coverage is asking: who audits the AI’s risk classification? Agentic Codeowners auto-approves PRs scored as low-risk across blast radius, complexity, and infrastructure impact. A supply chain attack disguised as a routine dependency bump would score low on all three dimensions — the exact profile the system is designed to wave through.
The most important thing Cursor shipped on March 5 wasn’t an AI feature. It was the moment agentic coding stopped being something developers do and became something that happens to codebases while developers sleep. The consumption flywheel, the MCP integrations, the event-driven triggers — they all point toward a future where most commits are authored by agents responding to automated signals, not humans responding to Jira tickets. The first security incident traced to an AI-auto-approved PR will decide whether that future arrives with guardrails or without them.
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.
