Cursor 3 Ships the Agent-First IDE. Worktrees, Best-of-N, and Design Mode Change the Game.

Cursor 3 launched April 2 and it's the biggest architectural change since the editor shipped. The IDE is now centered on an Agents Window for running many agents in parallel, across repos, locally, in worktrees, or in the cloud. This isn't a feature update. It's a rethink of what an IDE is.

Three features matter most. /worktree creates a separate git worktree so AI changes happen in isolation. Your main branch stays clean while the agent experiments. I've been doing this manually for months with Claude Code, creating worktrees by hand and spawning sessions in each one. Cursor just made it a slash command.

/best-of-n runs the same task across multiple models in parallel worktrees, then lets you compare outcomes. Want to see how Claude, GPT-4o, and Gemini each approach a tricky refactor? Run all three, diff the results, pick the best one. This is the first time I've seen a mainstream IDE treat model selection as a first-class workflow, not a settings dropdown.

Design Mode lets you annotate browser UI elements directly to guide agents. Click on a button, say "make this bigger," and the agent knows exactly what you mean. This solves the core vibe coding UX problem: describing visual changes in text is slow and imprecise. Direct manipulation is faster.

The timing with the Claude Code regression story is interesting. Cursor 3's /best-of-n gives you a hedge against any single model degrading. If Claude has a bad week, you still get work done because you're comparing outputs. That's a design decision that ages well in a world where model quality fluctuates.

What builders should do: try /best-of-n on your next non-trivial refactor. The comparison view alone will teach you things about how different models reason. And if you're managing a team, the Agents Window's parallel execution model means one developer can have 5-8 agents working simultaneously across different parts of a codebase. That changes how you plan sprints.