Cursor 2.0 turned the editor into a development environment and the parallel agents are the headline feature

The parallel multiple AI agents running simultaneously is the capability I had not seen anyone building toward before. Running separate agents on different parts of a codebase at the same time is not just faster, it is a fundamentally different development model where the bottleneck shifts from generation speed to review and integration of parallel outputs.

The Composer 1 model with a 200,000-token context window covers most real project codebases without running into limits. The integrated web browser with Chrome DevTools inside the editor removes the context switch between coding and testing that was previously unavoidable for frontend work.

The one-click screenshots and A/B testing UI redesigns as demonstrated capabilities are the things I want to test on a real client project. Has anyone used the parallel agents feature on a production codebase? What was the review process for integrating parallel outputs?