Cursor Hits $2B ARR and Ships Agentic Automations — What's Actually New
By EndOfCoding
Cursor just announced it has crossed $2 billion in annual recurring revenue — a milestone that took Figma 8 years and Cursor less than 3. But the bigger news isn't the number: it's what shipped alongside it. Cursor's new Automations system and self-hosted cloud agents fundamentally change what you can do with the tool. Here's a breakdown of what's actually new, how it compares to Claude Code and GitHub Copilot, and what it means for your workflow.
What You'll Learn
You'll understand Cursor's Automations system and how it differs from its previous agentic mode, what self-hosted cloud agents enable for teams, how Cursor's $2B ARR milestone stacks up in the AI coding tool market, and which new features are worth adopting now.
The $2B ARR Context
Cursor's trajectory:
- $0 to $100M ARR: ~18 months (faster than Notion, Figma)
- $100M to $1B ARR: ~8 months
- $1B to $2B ARR: ~4 months (accelerating)
For comparison: GitHub Copilot launched in 2021 and reached ~$200M ARR before Microsoft stopped reporting it separately. Cursor at $2B ARR is now larger than most standalone developer tools in history at this stage.
The driver: 73% of developers use AI coding tools daily (up from 41% in early 2025), and Cursor is the dominant choice for developers who want IDE-native AI. The market is expanding, not splitting.
The New Automations System
Cursor's previous agentic mode was session-based: you started an agent task, it ran, and it ended when the session closed. Automations change the architecture:
What Automations adds:
- Persistent background agents: Agents continue running between sessions — you can start a task, close Cursor, and come back to find it completed or in a defined waiting state
- Scheduled triggers: Run agents on a cron schedule ("every night at 2am, run the test suite and open a PR for any failures")
- Event-based triggers: Agents wake up on git events, file changes, PR opens, or CI failures
- Agent chaining: Output from one agent feeds into another — build a pipeline entirely within Cursor
Practical examples now possible:
# Automation 1: Nightly dependency update
Trigger: Cron (weeknight 2am)
Task: Run 'npm outdated', open PRs for patch-level updates
Approval: Auto-merge minor, flag major for human review
# Automation 2: PR security review
Trigger: Pull request opened (GitHub webhook)
Task: Review changed files for OWASP Top 10 patterns
Output: Comment on PR with findings and suggested fixes
# Automation 3: Test failure triage
Trigger: CI failure (GitHub Actions webhook)
Task: Identify failing tests, find likely cause, propose fix
Output: Draft PR with proposed fix + explanation
Self-Hosted Cloud Agents
Previously, Cursor agents ran on Cursor's cloud infrastructure — which raised questions for teams with IP-sensitive codebases. Self-hosted cloud agents allow you to:
- Run Cursor agents on your own infrastructure (AWS, GCP, Azure, or on-prem)
- Agent execution never leaves your VPC
- Full audit log of every tool call and file access
- Configure which Claude/GPT-4/Gemini models the agents use
- Compliance-friendly for SOC 2, HIPAA, and FedRAMP environments
This is enterprise-critical for financial services, healthcare, and government teams that need to use agentic coding tools but can't send code to third-party clouds.
How It Compares to Alternatives
| Feature | Cursor Automations | Claude Code | GitHub Copilot Agent |
|---|---|---|---|
| Persistent background agents | Yes | No (session-based) | Limited |
| Scheduled triggers | Yes | No | No |
| Event-based triggers | Yes (webhook) | No | Via Actions |
| Self-hosted execution | Yes (new) | Partial (API only) | Enterprise only |
| Parallel agents | Up to 8 | 1 per session | 1 per session |
| Model flexibility | Multi-model | Claude only | GPT-4o/Copilot |
The Multi-Model Strategy
Cursor's Automations support routing different tasks to different models:
- Complex architectural decisions → Claude 3.7 Sonnet or GPT-4o
- Code generation → Cursor's own small fine-tuned model (fastest, cheapest)
- Security review → Claude 3.7 Sonnet (best for reasoning-heavy tasks)
- Documentation → GPT-4o-mini (cost-optimized)
This model routing is configurable per Automation, not just per session.
What to Try First
If you're a current Cursor subscriber, Automations are available in Cursor 0.47+. Start with:
- The nightly dependency PR bot — set it up once, forget about manual dependency updates
- The pre-commit security scan — trigger on every commit, comment inline on vulnerabilities before code reaches PR review
- The test failure triage agent — hook into your CI, get AI triage before you even open the failing build
All three are in the Automations template library in the Cursor settings.
Common Challenges
'I use Claude Code — should I switch to Cursor?': They're increasingly complementary rather than competitive. Claude Code excels at complex, multi-file architectural tasks within a session. Cursor Automations excel at recurring, event-driven workflows. Many serious developers use both.
'Self-hosted sounds expensive to set up': The self-hosted option requires Cursor Business or Enterprise tier. Setup is Kubernetes-based and takes a few hours for a DevOps-capable team. For teams currently blocked from using agentic tools due to compliance, the setup cost is worth it.
'Are Automations GA or still in preview?': As of the March 25 announcement, Automations are GA for Cursor Pro and Business plans. Self-hosted is in limited availability — contact Cursor for access.
Advanced Tips
The Agentic Engineering angle: Karpathy's 'Agentic Engineering' framing (see our recent post) maps directly onto Cursor Automations. Designing an Automation is exactly the 'architect the system, direct the agents' model he's describing. You define the trigger, the task scope, and the approval gates — the agent executes.
Security note for self-hosted: Self-hosted agents execute shell commands on your infrastructure. Review the Cursor Automations permission model carefully — the same IDOR and path traversal risks that apply to any AI agent apply here. Least-privilege execution environments are non-negotiable.
The parallel agents productivity math: Running 8 parallel Cursor agents at $20/month Cursor Pro means each agent costs $2.50/month. If each saves 2 hours of developer time per month, that's an ROI that compounds. The Automations system is where that math starts to make sense at scale.
Deeper coverage in the ebook: The updated Vibe Coding Ebook Chapter 5: Tool Landscape includes Cursor's Automations and the updated $2B ARR milestone in the tool comparison section. Chapter 6: The Agent Revolution covers the parallel agent workflow in detail.
Stay updated: The Vibe Coding Academy intermediate track is being updated with an Automations module. Sign up to be notified when it launches.
Conclusion
Cursor's $2B ARR is impressive, but the real story is the Automations system — persistent, event-driven, multi-model agents that run on your own infrastructure. This is what the 'AI coding tool' category looks like at maturity: not just autocomplete, but programmable agent workflows that run while you sleep.
For the full tool comparison matrix and Cursor Automations deep dive, see Chapter 5 of the Vibe Coding Ebook. For weekly AI coding tool updates, subscribe at EndOfCoding.