Cursor Projects, launched September 10, 2026, is a persistent, cloud-based AI agent workspace where a single coordinator agent talks to you and delegates to a fleet of subagents that keep running whether your laptop is open or not. It maintains shared context over months of work, runs in the cloud by default, and can perform recurring work on a schedule. For product managers, the real shift is a shrinking PM-to-engineering handoff gap: a PRD can become an executable task breakdown, and feasibility questions get grounded answers in minutes instead of days.
What launched, how the coordinator + fleet model actually works, what it costs, and what it means for PM–engineering collaboration.
Cursor Projects is the first coding-agent product that genuinely changes the PM–engineering boundary — not because PMs will drive migrations themselves, but because a PRD can now become a grounded, executable task list without a week of back-and-forth. It's beta, it's infrastructure (not a teammate), and cost control is a real concern. But if your engineering team uses Cursor, this is the most important agent launch of 2026 to understand.
If you've used Cursor, you know the shape of a session: open a chat, describe a task, the agent edits files, you review, you move on. Close the tab and that context is gone. For a 20-minute fix, that's fine. For a three-week migration, you end up re-explaining your codebase to a fresh agent over and over.
Projects is built for the second case. Cursor describes it as a way to "take on larger bodies of work, such as a feature, a migration, or a full app," where the Project "maintains context over months of work" and "performs recurring work without being prompted." It's the concrete product behind the "third era of software development" vision Cursor floated in February 2026 — the one where you direct a fleet of agents instead of babysitting a single chat.
Cursor says it ran Projects internally for months before launch, including migrations of a few hundred pull requests and shipping Projects itself with a Project. Its headline figure: new users merge 30% more PRs, and heavy Projects users merge six times as many. That's a self-reported, internal stat — treat it as you would any vendor productivity number — but it tells you what the product optimizes for: throughput of merged work, not lines typed.
The real shift isn't "persistent chat" — it's that you stop talking to one agent and start talking to a coordinator that runs a fleet. You describe the goal; the coordinator breaks it down and dispatches specialized subagents that work in parallel. You review outcomes, not every keystroke.
The fleet runs in the cloud by default, not tied to your open editor — so subagents keep going whether your laptop is open or not. You get notified when there's something to review. That's the difference between "an AI that helps while I watch" and "an AI that makes progress while I'm in a roadmap review."
A Project holds context — the codebase, your decisions, prior subagent runs — across months. New contributors, human or agent, ramp without re-explaining the codebase. For a PM, that's the quiet win: the fifth time you ask "is this feasible given our auth model?" the Project already knows your auth model.
Subscriptions let a Project perform recurring work on a schedule without being prompted — dependency bumps, migration sweeps, test maintenance. It's the autopilot layer, and it's the part that frees engineering capacity for the feature work PMs are actually prioritizing.
This is the section that matters for product managers. Projects doesn't just make engineering faster — it moves the boundary where PM work ends and engineering work begins. Four concrete shifts:
PM writes a PRD; engineering translates it into tasks weeks later, losing context.
PM drops the PRD into a Project; the coordinator generates a task breakdown engineering can execute immediately.
PM asks "is this feasible?" and waits days for an engineer to spike it.
PM points a Project at the codebase and gets a grounded feasibility read in minutes from the Researcher subagent.
Context lives in one engineer's head and a stale doc; every handoff re-explains the codebase.
The Project's shared context persists across months — new contributors (human or agent) ramp without a re-explanation tax.
Recurring work (dependency bumps, migration sweeps) competes with feature work for eng capacity.
Subscriptions run recurring work on a schedule without being prompted, freeing eng for the work PMs are prioritizing.
The caveat: this only pays off if your engineering team actually adopts Cursor. If they're on a different stack, Projects is a curiosity, not a workflow. Pair this with our broader Cursor review for product managers and our guide on how to use Cursor for product management.
Projects is a feature rolling into existing Cursor plans during beta — there's no separate Projects price yet. The realistic floor is Pro, because the coordinator and its fleet burn through requests fast.
| Plan | Price | What it includes for Projects |
|---|---|---|
| Hobby (Free) | $0 | 2,000 code completions and 50 slow premium requests per month. Projects is rolling out to all users during beta, but on Hobby you'll hit the slow-request queue fast — fine for a tiny evaluation, painful for a multi-week project with a fleet of subagents running in the cloud. |
| Pro | $20/mo | Unlimited fast premium requests (fair-use cap), 500 fast premium requests for the most capable models, access to Claude 3.5 Sonnet and GPT-4o-class models. The realistic floor for using Projects — the coordinator and its subagents burn through requests, and slow-mode on Hobby makes the fleet crawl. |
| Ultra | $40/mo | Everything in Pro, plus 20x the fast premium request allowance and higher rate limits on the most capable models. The tier for anyone running Projects with a large fleet of subagents in parallel — heavy Projects users are exactly who this tier is built for. |
| Business | $40/user/mo | Everything in Pro, plus admin dashboard, team billing, SSO, privacy mode (no code retention), and centralized settings. The tier for PM + engineering teams adopting Projects together — privacy mode and admin controls matter when subagents are reading your codebase in the cloud. |
Prices reflect publicly listed Cursor tiers as of September 2026. Projects is in beta and rolling out; cloud compute usage may be metered as the product matures. Verify on the Cursor Projects announcement.
Codeium's AI editor with the Cascade agent for multi-step tasks. No persistent Project thread or fleet of subagents yet, but cheaper flat-rate pricing and strong for single-session work. See our Cursor vs Windsurf comparison.
Inline completions and chat inside your existing editor. Lighter on codebase-wide context and no coordinator/fleet model, but no new editor to install and the GitHub ecosystem integration is unmatched.
Anthropic's terminal agent for multi-file changes. No visual Project thread, but exceptional at long-context codebase reasoning — a strong complement, not a replacement, for a PM who lives in the terminal.
See the full ranked list in our Cursor alternatives guide, and the head-to-head in our Cursor vs Windsurf comparison.
50+ AI tools categorized by PM use case — prototyping, evaluation, research, analytics, roadmapping, and more. Each with a one-sentence description, a why-it-matters-for-PMs line, and a link. Free PDF, no credit card required.
Cursor Projects is a beta product moving fast — the coordinator model, the subagent fleet, and pricing will all shift in the coming months. This page captures the launch as of September 2026, but it can't tell you what shipped yesterday.
Signal Brief is the daily curation layer. Every weekday morning you get 5 new, vetted AI tools — each with a one-line description, a why-it-matters-for-PMs line, and a link. When a coding-agent launch like Projects lands, you'll read about it in your morning briefing, not three months later.
A one-time deep dive on the Projects launch. Read when you're evaluating the coordinator + fleet model.
5 new tools every weekday morning. 7-day free trial (25 tools), then $10/month or $96/year. The review is the map; Signal Brief keeps it updated.
Get 5 new AI tools for PMs every weekday morning. 25 tools during your free trial. Then $10/month or $96/year. Cancel anytime.
No credit card for the first 7 days. Cancel anytime.
Cursor Projects is a persistent, cloud-based AI agent workspace that launched on September 10, 2026. Instead of a single chat that forgets everything when you close the tab, a Project maintains context over months of work. A coordinator agent talks to you and delegates to a fleet of subagents that keep running in the cloud whether your laptop is open or not. It's built for larger bodies of work — a feature, a migration, or a full app — rather than a 20-minute fix.
Cursor Projects launched on September 10, 2026, and is rolling out to all users in beta. Cursor had run Projects internally for months before the public launch, including migrations of a few hundred pull requests and shipping Projects itself using a Project.
You talk to one coordinator agent. It breaks your request into tasks and farms each out to specialized subagents — a Planner, Coder, Reviewer, Tester, and Researcher — that run in parallel in the cloud. You don't babysit each agent; the coordinator orchestrates the fleet, reports back, and keeps going. Heavy Projects users merge roughly six times as many PRs as new users, per Cursor's own (self-reported) internal figures.
Partially. The biggest PM wins are indirect: faster, better-grounded feasibility reads and a PRD-to-task path that shrinks the PM–engineering handoff gap. A non-coding PM won't drive a migration Project themselves, but pointing a Project at the codebase to validate scope, or dropping a PRD in for a task breakdown, is genuinely useful. For hands-on coding PM work, see our broader Cursor review for product managers.
Projects is a feature rolling into existing Cursor plans during beta — there's no separate Projects price yet. The realistic floor is Pro at $20/mo, because the coordinator and its subagents burn through requests and Hobby's slow-mode makes the fleet crawl. Ultra at $40/mo is built for heavy fleet usage, and Business at $40/user/mo adds SSO, privacy mode, and admin controls for teams. Cloud compute usage may be metered as the beta matures.
Regular Agent mode runs one agent in your current session — close the tab and the context is gone. Projects is persistent: context survives across months, work runs in the cloud by default (not tied to your open editor), one coordinator directs many subagents instead of you driving one, and Subscriptions let a Project perform recurring work on a schedule without being prompted.
It's infrastructure, not a teammate — it doesn't replace product judgment, stakeholder management, or the messy human coordination of a real migration. It's in beta, so expect rough edges and self-reported (not independent) performance numbers. Subagents can escalate to expensive high-tier models without explicit consent, so cost control is a real concern. And non-developer PMs will hit a ceiling on what they can drive directly.
5 new AI tools every weekday morning. 7-day free trial, then $10/month or $96/year. Cancel anytime. No sponsorships, ever.
Questions? Email [email protected]