Cursor launched Projects in beta on September 10, 2026, rolling it out to all users the same day. Cursor describes Projects as built for work larger than a single pull request - a feature, a migration, or a full application - that needs to keep context over months rather than a single session. The product sits on top of a coordinator agent that, in Cursor’s words, “doesn’t write code itself but directs other agents that do,” fanning work out to what the company describes as thousands of parallel subagents running on dedicated cloud computers rather than the user’s laptop.
Three mechanics distinguish Projects from a conventional coding assistant. First, persistence: each Project keeps a set of shared files synced across machines, and agents record what they learn about a codebase and the user’s preferences so later agents on the same Project start more capable than the first one did. Second, reactivity: the coordinator can watch Slack channels, run on a fixed schedule, or track pull requests, and act autonomously on signals rather than waiting to be invoked. Third, delegation depth: the coordinator is explicitly not blocked by any single subagent’s work, so it can keep managing the Project while individual subagents run in parallel, with local agents spinning up only when needed for tasks like running tests.
Cursor’s own adoption numbers, published in the launch post, are the most concrete evidence offered for impact: new users who adopt Projects merge 30 percent more pull requests, and users who make Projects their primary workflow merge six times as many pull requests as before. Getting started is deliberately lightweight - a user opens Projects from the left navigation, describes an objective in natural language, and the coordinator takes it from there.
Projects is a significant step in Cursor’s positioning shift from pair-programr to what the company is effectively calling an autonomous engineering manager - a framing that matters given Cursor is now owned by SpaceX following that acquisition’s close earlier in 2026. The self-reported PR-merge multipliers are a useful signal of engagement but are not independently verified, and “six times as many PRs merged” says nothing about the quality or review burden of those PRs. What it does confirm is that persistent, multi-agent, cloud-native coding workflows are now a shipped product at a major IDE vendor, not a research demo.