Run Parallel Agents and Compare Model Outputs with Cursor 3
Cursor · New Product Launch · · major
Briefing for: Engineering
What happened
Cursor 3 introduces a dedicated Agents Window that supports running multiple AI agents in parallel across local, remote SSH, and cloud environments. New editor commands include `/worktree` for isolated git changes and `/best-of-n`, which runs the same task across multiple models simultaneously to compare results.
Why it matters
This shift from single-chat interactions to parallel orchestration significantly reduces idle time during long-running AI tasks. The ability to compare model outputs in real-time via worktrees allows for faster benchmarking of which LLM is best suited for specific complex logic or refactoring tasks.
What this enables
- If you are unsure which model to use for a task, use `/best-of-n` to see competing implementations side-by-side in isolated worktrees.
- If you work on complex features, use the new Agents Window to manage background tasks across remote environments without blocking your main editor.
- If you need to keep your git history clean, use `/worktree` to ensure agent-driven changes happen in isolation before merging.
Get personalized AI briefings for your role at Changecast →