Cursor Ships Projects Beta With Delegating Coordinator Agent
TL;DR
- Heavy Projects users merge 6x more PRs and new users merge 30% more, making human review capacity the new throughput ceiling.
- The coordinator plans and routes work to subagents but does not write code itself, keeping the human as final integrator.
- Cursor blocked autonomous pushes to main branches, choosing trust-building over raw capability as the primary adoption constraint.
Cursor released Projects in beta on September 10, extending its cloud-agent lineup with a supervisory "coordinator" that hands work off to thousands of subagents running in parallel. Authored by Alexi Robbins and Fredrika Lindh, the post reframes Cursor from a pair-programming editor into a manager-of-agents platform, with the coordinator itself staying out of the code. "The coordinator doesn't write code itself but directs other agents that do," they write, arguing that because it delegates rather than executes it is "never blocked and is always responsive to direction."
Projects run on dedicated cloud machines, keep files and research synced across local and remote sessions, and expose a subscription surface Cursor pitches as autonomous: "The coordinator can watch a Slack channel, run on a schedule, or follow all your PRs." The post walks through three use cases — building features, running migrations across hundreds of PRs, and ongoing "gardening" like lint-rule generation and component extraction. On one internal design-system Project, Cursor says the work "is on track to touch 20 to 100 PRs a day, so the coordinator organizes the work and the engineer checks in where attention is needed."
The productivity claims are Cursor's own and unaudited: new users merge 30% more PRs, and users who lean primarily on Projects merge six times as many. The post publishes no baseline, cohort size, or time window for either number.
Projects arrives on a Cursor cadence that's been running hot — three weeks ago the company gave cloud agents subscriptions, /goal and subagent VMs, and it lands the same day OpenAI shipped its Agents API in public beta, tightening the competition around managed multi-agent harnesses. It's our 19th Cursor story in the last 90 days, one of several recent items the coding-tools community has been circulating.
What others are reporting
-
AlphaSignal Read →
Identifies three primary use-case patterns (features, migrations, gardening) and flags that Projects fails on quick one-shots; compounding context-drift risk makes CI discipline a prerequisite.
new users merge 30% more PRs, while users who primarily use Projects merge six times as many
-
explainx.ai Read →
Frames Projects alongside the Sept 3 self-hosted infrastructure release as a two-layer system: one solving compute placement, the other solving orchestration memory and multi-week context.
You shouldn't have to onboard an agent every time you start a task.
-
ByteIota Read →
Highlights Cursor's deliberate choice to block autonomous main-branch pushes as a trust-building decision and adds practical spend-limit guidance for early adopters.
A Project runs on its own computer in the cloud, so closing your laptop doesn't stop it.
-
AlternativeTo Read →
Adds infrastructure specifics: cloud-computer deployment with optional local agent spinup for testing, plus Slack, schedule, and PR-event trigger integrations for autonomous task initiation.
Uses a coordinator agent to plan tasks and delegate implementation to parallel subagents.
Originally reported by cursor.com
Read the original article →Original headline: Cursor Ships Projects in Beta With Coordinator Agent That Delegates to Thousands of Subagents