Continually improving our agent harness

Keeping the Cursor app stable

Better AI models enable more ambitious work

Speeding up GPU kernels by 38% with a multi-agent system

Better MoE model inference with warp decode

A technical report on Composer 2

Improving Composer through real-time RL

Fast regex search: indexing text for agent tools

Introducing Composer 2

Training Composer for longer horizons

How we compare model quality in Cursor

Implementing a secure sandbox for local agents