Overview
Cursor is a fork of VS Code with AI baked in at every level: file-aware chat, multi-file edits, agent mode, tab-completion tuned on your codebase. For developers who spend more than a few hours a week in an editor, it's the biggest single productivity change of the AI era — and the pattern most other IDEs are now chasing.
Capabilities
- Tab-completion trained on the current file and open files.
- Cmd-K inline edits and Cmd-L codebase chat.
- Agent mode for multi-file changes with approval gates.
- Model choice: Claude Sonnet / Opus, GPT-4o, Gemini, and Cursor's own tab models.
- Rules for AI — repo-level .cursorrules file that sets conventions across sessions.
Where it shines
- Best-in-class inline edit UX; Cmd-K is muscle memory within a day.
- .cursorrules is a genuine team-productivity multiplier once tuned.
- Model routing means you get the best current model without switching tools.
Where it struggles
- Agent mode still needs supervision — see the plan-file refactor workflow.
- Large monorepos need careful indexing settings or costs balloon.
- Team pricing is per-seat and stacks fast at scale.
Best for
- Solo developers and small teams shipping full-stack apps.
- Anyone doing significant refactor work (agent mode + plan file).
- Codebases with strong conventions worth encoding in .cursorrules.
Pricing notes
Individual Pro is a no-brainer if you code more than 10 hours a week. Business tier adds SSO and admin — worth it above 10 developers.
Tutorials & patterns that work
Author a real .cursorrules file
Start with your team's actual style guide, not a generic 'be concise' rule. Encode file structure, naming conventions, error-handling patterns, and testing style. Commit it to the repo.
Use the plan-file pattern for refactors
See the Cursor plan-file refactor workflow. Ask for plan.md before code on any change touching more than 5 files. Prevents the sprawl that kills agent-mode ROI.
Route models by task
Tab-completion → Cursor's default. Cmd-K short edits → Sonnet. Multi-file refactor → Opus or Sonnet with agent mode. Long context / codebase Q&A → Gemini. Don't use one model for everything.