CursorEngineering PostOfficial Source

Introducing Projects

Cursor Projects gives a coordinator persistent context, cloud and local subagents, and event-driven work for features, migrations, and maintenance that outlive one chat.

Cursor · Sep 10, 2026

Open Source Open Markdown Open JSON

Source Summary

Cursor Projects is a beta workspace where a coordinator delegates to **thousands of subagents**, maintains shared context for **months**, and reacts to schedules, Slack, or PR events. Cloud agents can hand local testing to an agent on your machine.

Practical Implication

Use a Project for work spanning several PRs: research first, preserve discoveries as shared files, then let the coordinator partition implementation and tests. For migrations or recurring maintenance, begin with close review and relax oversight only after repeated fixes hold up.

Agent-Ready Context

Cursor Projects is a beta workspace where a coordinator delegates to **thousands of subagents**, maintains shared context for **months**, and reacts to schedules, Slack, or PR events. Cloud agents can hand local testing to an agent on your machine.

Use a Project for work spanning several PRs: research first, preserve discoveries as shared files, then let the coordinator partition implementation and tests. For migrations or recurring maintenance, begin with close review and relax oversight only after repeated fixes hold up.

Cursor reports new users merging **30% more PRs** and Project-heavy users merging **6× as many**, but these are internal usage figures rather than controlled results. Availability is still beta, and the post does not detail cost, permission boundaries, or failure rates at fleet scale.

Connected Context · Feed7 Judgment

Projects expands the shared cloud-workspace pattern into a long-lived coordinator that can partition multi-PR work across a very large agent fleet and react to external events. Compared with earlier background and channel-based agents, it raises the coordination ceiling substantially, but beta status and missing cost, permission, and failure data make close review essential before its scale or internal PR figures are treated as evidence of dependable autonomy.

Realtime multiplayer, automation, and you! — Idan Gazit, GitHubBoth move coding agents into persistent, team-context workflows triggered beyond personal chat; Projects adds explicit large-scale delegation and months-long shared context, while the GitHub prototypes emphasize declarative permissions and reviewable workflow boundaries.Vercel Agent is now available in Slack code channelsSlack can supply triggers and context to both approaches, but Vercel describes read-only defaults, approvals, and action records that are not detailed for Projects, sharpening the governance questions raised by its fleet scale.Firetiger joins CursorProjects supplies the long-running coordination substrate that could eventually act on monitoring events, while Firetiger extends Cursor’s intended loop into rollout checks and incident investigation; neither source establishes safe autonomous production remediation.openai/codexCodex unifies one agent across local and hosted surfaces, whereas Projects introduces a higher-level coordinator spanning many agents and handing local tests to a machine agent, making orchestration and shared context the differentiating layer.

Context Map

toolscoding#coding-agents#cloud-agents#multi-agent

Uncertainty

Cursor reports new users merging **30% more PRs** and Project-heavy users merging **6× as many**, but these are internal usage figures rather than controlled results. Availability is still beta, and the post does not detail cost, permission boundaries, or failure rates at fleet scale.