Cursor Cloud Agents: AI That Builds, Tests, and Records Video Demos on Its Own

What Are Cursor Cloud Agents?

Cursor Cloud Agents are fully autonomous AI coding agents that each run in their own isolated virtual machine with a complete development environment. Unlike inline copilots that suggest code while you type, Cloud Agents work independently — you describe what you want, and the agent builds it, tests it, and submits a pull request.

The Numbers That Matter

Internally at Cursor, 30% of merged PRs are now created entirely by Cloud Agents. This isn't a demo metric — these are production features, bug fixes, and refactors on Cursor's own codebase.

Bugbot Autofix, which monitors incoming PRs and automatically fixes issues, has hit a 35% merge rate. Overall bug resolution rates increased from 52% to 76% over six months.

How Cloud Agents Work

1. Isolated VM Environment

Each agent gets its own virtual machine with full access to the filesystem, terminal, and browser. The agent can:

Clone repositories and install dependencies
Run build and test commands
Launch a browser and navigate to localhost to verify visual output
Compare screenshots before and after changes
Record video of its entire workflow

2. The Review Workflow

Cloud Agents produce merge-ready pull requests with artifacts:

Code changes: Clean, well-structured diffs
Test results: Full test suite output
Video recording: Step-by-step replay of what the agent did
Screenshots: Before/after visual comparisons
Logs: Complete terminal output for debugging

3. Bugbot Autofix

Bugbot watches your PR pipeline and automatically fixes issues it detects. When a reviewer leaves a comment like "this should handle the null case," Bugbot generates a fix commit within minutes.

When to Use Cloud Agents

Great For:

Bug fixes with clear reproduction steps: Describe the bug, agent reproduces and fixes it
Feature implementation from specs: Provide a detailed spec, agent builds it
Refactoring tasks: "Convert this class component to a React hook" — agent handles the migration
Test generation: "Add unit tests for the auth module" — agent writes and runs them

Not Ready For:

Architectural decisions requiring deep product context
Security-critical code that needs human review regardless
Performance optimization requiring profiling and benchmarking

The Shift From Copilot to Agent

Cloud Agents represent a fundamental workflow shift. Instead of "AI helps me code," it becomes "AI codes and I review." The experience is closer to managing a junior developer than using a tool:

You assign tasks in natural language
The agent works independently for minutes to hours
You review the output, approve or request changes
The agent iterates based on your feedback

For teams already drowning in code review backlogs, this isn't just convenient — it's a force multiplier. One senior engineer can now effectively manage 5-10 Cloud Agents working in parallel on different tasks.

Getting Started

Ensure you have Cursor Pro or Business subscription
Enable Cloud Agents in Settings > Features
Open the Agent panel and describe your task
Review the PR when the agent completes

What This Means for the Industry

When 30% of a development tool company's own PRs come from AI agents, the signal is clear: autonomous coding is production-ready for well-defined tasks. The question isn't whether agents will write most code — it's how quickly teams adapt their workflows to the review-centric model.