Cursor 2.0: Composer, multi-agent UI, sandboxed terminals

Cursor 2.0 launches Composer (fast agentic coding), parallel multi-agent workflows, GA browser testing, improved reviews, and sandboxed terminals.

by HowAIWorks Team
CursorComposerAgentsIDESandboxBrowserCode ReviewProductivity

Introduction

Cursor announced a major release with Cursor 2.0, introducing its first coding model Composer and a redesigned, multi-agent interface. The update focuses on faster agentic coding loops, parallel execution without file conflicts, improved review/testing workflows, and enterprise-grade safety with sandboxed terminals (blog, changelog).

What’s new in Cursor 2.0

Composer: fast agentic coding model

  • Frontier coding model built for low-latency agent workflows
  • Reported to be ~4x faster vs similarly capable models, with most turns <30s
  • Trained to leverage codebase-wide semantic search for large projects

Multi-agents in parallel

  • Run up to eight agents in parallel on a single prompt
  • Isolation via git worktrees or remote machines to prevent file conflicts
  • Sidebar manages agents and plans in the new editor UI

Browser for Agent (GA)

  • In-editor browser tooling with element selection and DOM forwarding to agents
  • Enterprise support expanded in 2.0

Sandboxed terminals (GA on macOS)

  • Default sandbox execution for agent commands with workspace read/write and no internet by default
  • Admin controls for enterprise to enforce sandbox, git, and network access policies

Improved code review experience

  • Easier, consolidated diff review across multiple files without jumping between them

Team commands and rules

  • Define team-level custom commands and rules in the dashboard; centrally managed and auto-applied to teammates

Voice mode

  • Control Agent with speech-to-text; configurable submit keywords

Performance upgrades

  • Faster Python and TypeScript LSPs on large projects with dynamic memory limits
  • Memory leak fixes and improved overall responsiveness, especially when viewing diffs

Plan Mode in background

  • Plan with one model, build with another; optionally run plans in parallel to compare

Prompt UI and harness improvements

  • Inline file/directory pills, better copy/paste of tagged context
  • Agent harness upgraded across models, with notable gains for GPT-5 Codex

Cloud agents and enterprise features

  • 99.9% reliability, instant startup for cloud agents; improved send-to-cloud flow
  • Enterprise hooks distribution from web dashboard and audit logs

Why it matters

Cursor 2.0 targets the two biggest bottlenecks emerging with agentic coding—review and testing—while accelerating iteration speed. Parallel agents, better diffs, and a native browser tool aim to shorten the loop from planning to validated changes, especially on large TypeScript/Python codebases. Sandboxed execution and enterprise controls improve safety and governance.

Availability

Cursor 2.0 is available now. You can download it via the official site and review the full changelog for details (blog, changelog).

Sources


Explore more AI development tools in our AI Tools catalog and model updates in Models.

Continue Your AI Journey

Explore our lessons and glossary to deepen your understanding.