Cursor

AI-first code editor. $2B ARR in 24 months — fastest scaling in B2B SaaS history. Pioneering RL-from-user-feedback for coding agents.

Key Technical Contributions

• Real-Time RL for Composer: New checkpoint every ~5 hours from user feedback
• Self-Summarization: RL-trained self-compression for long agent contexts
• Warp Decode: 1.84x MoE inference speedup on Blackwell via output-centric parallelism
• CursorBench: Internal eval suite using real session data
• Online RL for Tab: 15% improvement from implicit user feedback
• Codebase indexing as core technical moat

Business Context

• $100M\to$2B ARR in under 24 months
• Central example in token economy bubble analysis
• Competitive dynamics with Cognition (embeddings vs grep architectures)

Connections

• Coding Agents — leading example
• RL Infrastructure — production RL from user feedback
• Reward Hacking — discovered reward hacking in Composer RL