Cursor Launches Beta of AI Agent Management Platform 'Projects'

Cursor has launched the beta version of the Projects feature, which allows users to manage large-scale development tasks and maintain context over several months. This feature materializes the vision of the "Third Era of Software Development" announced in February, designed to reduce the burden of agent management and allow users to focus on task instructions.

The core architecture is the Coordinator agent, which does not write code directly but only directs other agents. This ensures the system remains always responsive without bottlenecks. Internal testing results showed a 30% increase in PR merge rates for new users and a 6x increase in merged PRs for key users, demonstrating significant productivity gains.

3 Core Features of Projects

• Cloud-first execution: Projects run on their own computers, persisting regardless of laptop shutdowns. The Coordinator calls local agents only when local testing is required.
• Shared context: All agents share files and context. Research results and codebase learnings accumulate over time, improving the Coordinator's efficiency.
• Subscriptions: Signal-based automated execution can be set up for tasks such as monitoring Slack channels, scheduling, and PR tracking, allowing work to be processed without prompt input.

Key Usage Patterns

Internally, Cursor utilizes Projects in three patterns: Feature work, Migrations, and Gardening. In particular, in the Gardening pattern, the Coordinator scans new PRs, detects repetitive mistakes, and automatically adds lint rules, with reports of autonomously handling 20–100 PRs per day. Projects are currently being rolled out in beta to all users.