You ask your AI coding assistant to build a feature. It generates code, but you're not sure if it's the best approach. What if you could run eight different AI agents on the same task simultaneously, compare their solutions, and pick the best one?
Cursor 2.0 launched this week with exactly that capability. The AI code editor now lets developers run multiple agents in parallel, includes its own custom coding model, and shifts the entire interface from file-centric to agent-centric workflows.
What Cursor 2.0 Actually Does
Version 2.0 introduces several major changes to how developers interact with AI during coding.
Parallel Agent Workflows: Run up to 8 agents simultaneously on the same prompt using isolated git worktrees. Each agent works in its own copy of your codebase to avoid conflicts.
Composer Model: Cursor's first custom coding model, trained specifically for agentic coding tasks. It completes most iterations in under 30 seconds and generates code at 250 tokens per second, roughly 4x faster.
Agent-Centric Interface: The UI now centers around agents and their plans rather than files. You see what agents are doing, not just which files are open.
Built-In Browser: Embedded browser lets agents select UI elements, test interfaces, and validate changes without leaving the editor.
Security Features: Sandboxed terminals on macOS execute non-allowlisted shell commands with restricted networking and file access.
Limitations
- Significant functionality removed: Version 2.0 removed features like past chat history and specific git commit context. Some users report it feels like a step backward.
- Performance inconsistencies: Multiple users report less intelligent responses, frequent mid-response cutoffs, and loss of multi-step execution compared to version 1.7.
- Codebase understanding issues: Some developers report the tool no longer grasps entire codebases well, making random modifications that break existing functionality.
- Usage costs: The credit-based system can deplete quickly on heavy tasks. Multi-file edits and advanced reasoning consume credits faster.
Pricing Structure
Cursor offers three tiers:
- Free: Unlimited basic completions with capped premium requests
- Pro: $20/month with unlimited tab completions, unlimited "Auto" model usage, and $20 worth of credits for premium models
- Business: $40/user/month adding team features, SSO, admin controls
Credits deplete based on model API costs. Heavier models, longer contexts, and advanced features consume credits faster. When credits run out, you can switch to the Auto model or enable pay-as-you-go overages at API rates.
Our Assessment
For new users evaluating AI coding tools, Cursor 2.0's approach is worth trying, but be prepared for a learning curve and potential frustrations as the platform matures.
The real value emerges when you have complex problems with multiple valid solutions. Being able to run parallel agents and compare approaches could save hours of trial and error. Just don't expect it to replace careful code review and testing.
See you next week with more 10x hacks!
