Users who primarily use Cursor’s new Projects feature are merging six times as many pull requests as before. That single stat tells you more about what Cursor is attempting here than any feature list could. This isn’t another chat improvement or context window bump. It’s a structural shift in how AI-assisted development works.
Cursor has announced Projects, now in beta and rolling out to all users. The feature is built around a coordinator agent that doesn’t write code itself but directs other agents that do. You talk to the coordinator, the coordinator delegates to subagents, and the work gets done, even while your laptop is closed.
What Projects actually does
The coordinator model is the key architectural decision here. Because the coordinator delegates rather than executes, it’s never blocked by a single long-running task. It can spin up subagents in parallel, across cloud machines and locally when needed, and stay responsive to your direction throughout.
Three capabilities make the whole thing work:
- Cloud-first execution: Each Project runs on its own machine in the cloud. Your laptop going to sleep doesn’t pause it. Local agents get spun up only when something needs testing on your machine specifically.
- Shared context: Every agent working on a Project writes back to a shared set of files. If one agent figures out how to test a service, every future agent gets those instructions automatically. The coordinator gets more effective as the Project runs longer.
- Subscriptions: The coordinator can watch a Slack channel, follow your PRs, or run on a schedule. It acts on signals without waiting for you to prompt it, which is where a lot of the “while you’re away” value comes from.
How Cursor’s own engineers are using it
Cursor has been running Projects internally for months, and the three use cases they highlight map to real engineering pain points. Feature work is the most straightforward: a Project starts by researching the codebase, builds shared context, creates a plan, and sends agents to implement different parts in parallel. After shipping, the same Project monitors logs and handles bug reports with full knowledge of the original decisions.
Migrations are where Projects gets particularly interesting. Cursor used it internally to adopt new frameworks and replace styling systems across hundreds of PRs. The pattern is simple: you establish a safe approach with the coordinator, then it applies that approach incrementally. You review closely at first, then less and less as the fixes hold up.
The third pattern, which Cursor calls “gardening,” covers the maintenance work that never ends. One engineer on the team runs a design-system Project that scans every new PR, extracts components that belong in the design system, and adds lint rules when it sees the same mistake twice. That Project is on track to touch 20 to 100 PRs a day.
Why this matters beyond Cursor
The broader context here is that Cursor is making a direct bet on multi-agent orchestration as the next layer of developer tooling. GitHub Copilot, Codeium, and Replit are all pushing agentic features, but most are still operating at the single-agent, single-task level. A persistent coordinator that accumulates project-specific knowledge over months is a meaningfully different model.
The shared context system is worth watching closely. Institutional knowledge about a codebase has always been the thing that makes senior engineers valuable and new hires slow. If a coordinator agent can actually build and apply that knowledge reliably, the productivity math changes in ways that go well beyond PR counts.
Projects is available in beta starting today. You can start one from the left-hand nav, describe what you want built, and the coordinator handles the rest. It’s designed for work that will outlive a single chat session, which, honestly, is most of the interesting work.
