The reduction was achieved without degrading agent quality through tighter prompts, selective tool loading, better caching, and compressed file reads.
Cursor users consume fewer tokens per agent run while maintaining output quality.
Read this as text
Back to the AI news