Training Composer for longer horizons

By making self-summarization part of Composer's training, we can get training signal from trajectories much longer than the model's max context window.

How we compare model quality in Cursor

We use a hybrid online-offline eval process to keep our understanding of model quality aligned with what developers actually do.

Implementing a secure sandbox for local agents

How we built agent sandboxing on macOS, Linux, and Windows to reduce interruptions while improving security.

Introducing Composer 1.5

Improved reasoning over challenging coding tasks by scaling RL over 20x.

Towards self-driving codebases

We're making a part of our multi-agent research harness available to try today in preview.

Securely indexing large codebases

By securely reusing a teammate's existing index, we cut time-to-first-query from hours to seconds on the largest repos.

Building a better Bugbot

How we used a custom AI-driven metric to systematically improve Bugbot.

Scaling long-running autonomous coding

We've been experimenting with running coding agents autonomously for weeks at a time.

Dynamic context discovery

As models improve as agents, we've found success by providing fewer details up front, making it easier for the agent to pull relevant context on its own.

The productivity impact of coding agents

A new study from the University of Chicago finds that companies merge 39% more PRs after Cursor's agent became the default.