Cursor's new "Projects" agent keeps working after you close your laptop -- for months
A coordinator agent that doesn't write code itself, only plans and delegates to subagents, runs in the cloud and can trigger itself from Slack, a schedule, or an incoming PR -- a shift from IDE copilot toward an always-on agent product.
Cursor launched Projects on September 10, 2026, rolling out in beta to all users. A "coordinator" agent -- which doesn't write code itself -- plans a body of work and delegates pieces of it to subagents, running in the cloud so it keeps going after the developer closes their laptop. It can also trigger itself: from a Slack message, a schedule, or an incoming pull request, rather than only starting when someone opens the editor.
From assistant to always-on process
Per the company's own announcement, Projects lets a team "take on larger bodies of work, such as a feature, a migration" -- framing that marks a real shift from Cursor's original pitch (an AI-assisted editor) toward something closer to a standing engineering process that happens to be triggered by, rather than tethered to, a human sitting at a keyboard.
The cost question this raises
An agent that keeps running unattended, potentially for months, changes the usage-based cost math for anyone already budgeting Cursor seats: spend is no longer bounded by how long a person is actively typing. For engineering and finance leaders tracking AI-tool spend, "always-on and self-triggering" is the detail worth pricing out before rolling this out broadly, not just the feature list.