Cursor
AI-native code editor that has become the preferred IDE for developers operating at Levels 2–4. Past $500M ARR as of 2025, making it one of the fastest-growing developer tools in history. Integrates multiple AI models and supports agentic workflows within an IDE experience.
Cursor is the AI-native code editor that succeeded where GitHub Copilot stumbled at the workflow level. Where Copilot bolted AI onto an existing editor (VS Code), Cursor rebuilt the IDE around AI as a first-class citizen.
Why It Matters
For Dan Shapiro’s Levels 2–4, Cursor is the most practical daily driver. It handles:
- Tab completion: Context-aware multi-line completions
- Inline editing:
Cmd+Kto edit code in natural language - Chat: Ask questions about your codebase with full file context
- Composer/Agent: Multi-file changes, running commands, iterating on test failures
- Rules for AI: Per-project and per-directory AI instruction files (similar to CLAUDE.md)
Market Position
- $500M ARR as of 2025 — making it one of the fastest-growing dev tools ever
- Primary competitor to GitHub Copilot for developers who want serious multi-file capabilities
- Many teams use both: Cursor as their daily IDE and Claude Code for longer autonomous runs
The Level 4 Workflow
At Level 4 (spec → AI does implementation → human evaluates outcomes), Cursor’s Composer mode is the most polished implementation available. Engineers write a detailed description, Cursor proposes a plan, implements across files, runs tests, and iterates.
The difference from Level 5: the human is still present, reviewing the implementation plan and accepting/rejecting changes.
Limitations
- Still IDE-bound: works within a session, not fully headless/autonomous
- Large codebase performance requires careful context management
- No native external scenario testing methodology