Cursor launches Projects, letting a coordinator agent run months-long coding work across thousands of subagents

Cursor released Projects on September 10, 2026, a new mode built around a single coordinator agent that plans and delegates work to as many subagents as a job requires, rather than writing code itself. The feature targets work that spans weeks or months — a full feature build, a large migration, or an entire app — instead of the single-session tasks Cursor's agent mode was originally built for.

What's new

According to Cursor's announcement: "Projects lets you take on larger bodies of work, such as a feature, a migration, or a full app. It maintains context over months of work, delegates tasks to thousands of subagents."

The coordinator itself never touches code directly. Cursor writes: "The coordinator doesn't write code itself but directs other agents that do. Because it delegates rather than executes, it is never blocked and is always responsive to direction." That division of labor is the core design change from Cursor's existing agent mode, where a single agent both plans and implements.

Projects also persist state across sessions. The coordinator retains institutional knowledge about a codebase — prior decisions, conventions, and preferences — so it does not need to be re-briefed each time a developer returns to the work. Cursor is pairing this with a subscription tier that turns a Project into a standing operator: a coordinator can watch a Slack channel, run on a fixed schedule, or follow pull requests, fixing CI failures or responding automatically when a PR opens or merges. Because a Project runs on Cursor's own cloud infrastructure rather than a developer's laptop, that monitoring continues after the laptop is closed.

Context

The release follows two other agent-infrastructure moves from Cursor this month. On September 3, Cursor shipped the ability to run cloud agents on infrastructure a team already manages — AWS Lambda, Cloudflare, Coder, Daytona, E2B, Modal, Namespace, and Vercel — plus computer-use support for self-hosted workers on Linux and Mac, letting an agent click, type, screenshot, and drive a browser directly. Projects builds on that same cloud-execution base, adding a management layer for coordinating many agents at once rather than running one at a time.

The move also mirrors a broader shift among coding-agent vendors toward long-running, semi-autonomous work rather than single-prompt code generation. Competing tools have pushed in similar directions this year with persistent memory and background-agent features, and Cursor's own pricing and positioning have increasingly emphasized delegated, multi-agent workflows over single-session pair programming.

Why it matters

Projects is a bet that the bottleneck in AI-assisted coding is shifting from code generation quality to coordination: keeping dozens or hundreds of parallel agents pointed at a coherent goal without a human re-explaining context every time. If the coordinator model holds up on real, multi-week engineering work — not just demos — it could change how engineering teams staff big migrations, treating an agent swarm as a persistent team member rather than a tool invoked per-task. It also raises the same governance questions larger vendors are already grappling with: a coordinator that runs unattended on a schedule, watches Slack, and merges its own fixes needs permissioning and audit trails commensurate with a junior engineer with production access, not a code-completion plugin.