Cursor Launches Autonomous Cloud Agents
AI-native code editor Cursor has launched Cloud Agents, autonomous agents that operate on dedicated virtual machines to handle software development tasks. The agents can clone codebases, execute tests, record video demos of their work, and submit pull requests to GitHub. The system is designed to provide an auditable trail of an agent's actions before a human review.
- Cursor is dogfooding this technology heavily, with over 30% of pull requests merged internally being created autonomously by these cloud agents. - The key differentiator from other AI coding tools is the agent's ability to perform "computer use" in its sandboxed environment; it can run the application, navigate the UI as a user would, and visually verify changes, which helps catch bugs that unit tests might miss. - By moving agents from a developer's local machine to the cloud, Cursor allows for running many agents in parallel—potentially 10 to 20 at a time—each in its own isolated virtual machine without competing for local resources. - This launch follows a period of massive growth for Cursor's parent company, Anysphere, which was founded by four MIT graduates and reached a valuation of $29.3 billion after a $2.3 billion funding round in late 2025. - While fully autonomous agents like Devin often operate in a separate environment like Slack, Cursor's agents are designed to be triggered from and integrated directly within the developer's existing workflow, including the editor, GitHub, or Slack. - The Cloud Agents feature is available across Cursor's pricing tiers, which include a free "Hobby" plan, a $20/month "Pro" plan for individuals, and a $40/user/month "Business" plan for teams. - A core part of the workflow is the delivery of artifacts with each pull request; developers receive not just code, but also video recordings, screenshots, and logs that serve as evidence of the agent's work and successful tests.
