#414 Cursor

A quick introduction to Cursor, an AI-powered code editor.

Notes

Cursor is an AI-powered code editor developed by AnySphere, a startup founded by MIT graduates in 2022. It’s built as a fork of Visual Studio Code, combining the familiar VS Code interface with deeply integrated AI capabilities that go far beyond simple autocomplete.

There is a Hobby (Free) plan, with limited daily completions.

Product Architecture

Local IDE & CLI
- Agent mode: Complex features, refactoring; Autonomous exploration, multi-file edits; All tools enabled.
- Ask mode: Learning, planning, questions; Read-only exploration, no automatic changes; Search tools only.
- Plan mode: Complex features requiring planning; Creates detailed plans before execution, asks clarifying questions; All tools enabled.
- Debug mode: Tricky bugs, regressions; Hypothesis generation, log instrumentation, runtime analysis; All tools + debug server
Web-based UI for cloud agents - only paid plans

Where Cursor Excels

Large-Scale Refactoring
- Rename functions across entire repos, update imports, and apply changes consistently
- Example: “Rename generateCartTotal to calculateCartTotal across all files”
Complex, Multi-File Projects
- Navigate and modify interdependent codebases
- Better context awareness than Copilot for monorepos
AI-Driven Development (“Vibe Coding”)
- Natural language to working code: describe features, let the agent implement
- End-to-end task automation from planning to PR creation
Codebase Exploration & Onboarding
- Ask questions about unfamiliar codebases
- Get instant explanations with referenced source locations
Teams Needing Model Flexibility
- Switch between models based on task requirements
- Compare outputs from multiple models in parallel

Less Suitable For

Simple, single-file scripts where basic autocomplete suffices
Developers deeply embedded in GitHub-centric workflows who prefer minimal disruption
Budget-conscious individual users

Standout Features

Project-Wide Context Awareness
- Automatically indexes your entire codebase when you open a project
- Understands cross-file relationships and dependencies
- Enables multi-file refactoring with natural language commands
AI Code Completion (Tab)
- Suggests multi-line code blocks and entire functions
- Predicts your next edit location
- Auto-imports symbols for TypeScript/Python
Local, Workspace and Cloud Agents
- Complete end-to-end coding tasks without supervision
- Can write code, commit changes, and create pull requests autonomously
Multi-Model Support
- Choose from GPT-4o, o1, Claude 3.5/4.5 Sonnet, Gemini 2.5 Pro, Grok Code, and custom models
- Parallel agents: Run multiple models simultaneously on the same prompt, each in separate Git worktrees
Advanced Tooling
- MCP (Model Context Protocol) servers: Connect to external systems (Google Drive, Notion, etc.)
- Subagents: Spawn specialized agents for complex tasks
- Plan Mode: Generates detailed .MD plans with “Build” buttons to spawn new agents
Privacy Mode
- Zero data retention—code isn’t stored or used for model training
- Local file indexing until you initiate queries
- Enabled by default for team plans