Cursor's cloud agents now boot 3x faster

Cursor shipped “Builds” on August 13, a system that maintains warm copies of pre-configured development environments so cloud agents can boot directly into a working setup instead of provisioning one from scratch each run. Cursor says the approach cuts startup time by 10x internally and delivers roughly 3x faster time to first token in practice. Builds that fail to configure correctly don’t become active, so an agent keeps working from the last good environment while a developer debugs the failure in the background, rather than the whole run stalling. A new Builds tab shows status, logs, and the commit each build is based on.

The feature builds on Cursor’s push into cloud-hosted agent work generally, following its Google Workspace plugins earlier this month that let agents read and manage Drive, Gmail, and Calendar without leaving the editor, and its expansion onto iPad as a lighter-weight surface for reviewing and steering the same cloud agents.

What it means for you

Environment setup time is an unglamorous but real tax on any cloud-agent workflow, if you’re running agents against a monorepo or a dependency-heavy stack, a chunk of every task’s wall-clock time has likely been spent installing packages and configuring state rather than doing the work you asked for. Cursor’s fix, caching known-good environments and falling back safely when a build breaks, is a pattern worth checking for in whatever agent platform you use even if you’re not on Cursor specifically, and it pairs with OpenAI’s new Ultrafast inference tier as another sign that shaving latency out of the agent loop, not just improving model quality, is where a lot of this week’s competitive pressure is landing.