Cursor shipped Projects to all users starting September 10, 2026, and this is not an incremental feature update. This is Cursor redefining what a coding tool actually is. Projects moves the product from session-based AI assistance into persistent, multi-agent project orchestration. Your agents now live on their own cloud computer, run for months, manage fleets of subagents in parallel, and keep working after you close your laptop. If you lead an engineering team and you haven't looked at this yet, clear your afternoon.

Cursor Projects introduces three capabilities that together represent a fundamental architectural shift in how software gets built: Cloud-by-default execution. Each Project runs on its own dedicated cloud computer. A coordinator agent lives there permanently, maintaining context across months of work and spinning up thousands of subagents in parallel. When local testing is required, the coordinator reaches out and spins up a local agent on the developer's machine. The distinction matters: this is not a cloud sync of your IDE session. This is a separate compute environment that owns a long-lived slice of your engineering work. Shared context. Every agent interaction within a Project writes back to a shared context store: files, research, learned workflows, prior decisions. All agents in a Project read from the same memory. Over weeks and months, this becomes something genuinely new: a machine-maintained record of how your team builds software, which patterns you prefer, how your CI behaves, which migrations you've run. It is organizational memory that doesn't leave when an engineer quits. Subscriptions. The coordinator agent can subscribe to live signals: Slack channels, pull requests, CI failures, scheduled triggers. It acts without being explicitly prompted. You assign a feature or migration and the coordinator researches the system, creates a plan, delegates parallel implementation to subagents, and keeps CI green without you driving every step. This is the piece that separates Projects from anything else currently in the market. The workflow Cursor describes for features and migrations is worth understanding precisely:

That is a delivery loop. Not an autocomplete.

Cursor's annualized revenue grew from $100 million in January 2025 to more than $4 billion by June 2026. SpaceX agreed to acquire the company for $60 billion in stock. That number puzzled a lot of people when it was announced. Projects is the explanation. The acquisition was never really a bet on a code editor. It was a bet on AI-native software factories: persistent agent infrastructure that can own entire workstreams, accumulate organizational process knowledge, and operate at a scale that individual engineers cannot match. Projects is the product that justifies that thesis. Cursor is not competing with Copilot. It is competing with the idea of a human-staffed feature team. That reframes the competitive picture significantly. GitHub Copilot Workspace and Anthropic's Claude Code are both capable tools, but they operate within sessions. They assist engineers. Cursor Projects operates across months. It orchestrates engineers and agents together. The gap between those two models is not a product roadmap question; it is an architectural one.

Most coverage of Projects will focus on velocity. That is the obvious frame and it is correct as far as it goes. But the more durable competitive consequence is organizational memory.

Every time the coordinator researches your codebase, every preference it learns about how your team reviews PRs, every pattern it absorbs from watching your incident response: that becomes part of the shared context store. Over twelve months, a well-used Project accumulates a living playbook of your engineering practices. The coordinator learns that your team rolls back by reverting the migration before the service restart. It learns that compliance-related fixes require a second review. It learns which tests are flaky and which are load-bearing.

This is not a feature you demo. It is a moat that compounds. Teams that start using Projects now will have coordinators that are materially smarter about their specific systems in 2027 than teams that start later. The tool gets better the longer you use it, not just because the underlying models improve, but because your organizational context accumulates. This also raises real governance questions, which engineering leaders need to address directly rather than deferring.

Akeyless released a Cursor extension on September 10 specifically targeting AI workflow security, citing research that AI-assisted code leaks secrets at roughly double the rate of the overall GitHub baseline. That number should be in your briefing to your CISO before you pilot Projects at any meaningful scale.

The elevated leak rate is not a Cursor-specific failure. It reflects a general pattern: agents that move quickly across many files, spin up new configurations, and write infrastructure code are handling secrets-adjacent material at much higher frequency than a developer in a single session. The mitigation is not to avoid AI agents; it is to instrument them. Akeyless is building in the right direction. Your secrets management tooling needs to be in place before you run a coordinator agent against your production codebase.

Netlify's move to support Cursor Origin as a Git provider on the same day as the Projects launch signals how quickly the ecosystem is organizing around Cursor's infrastructure. Deployment pipelines, secrets management, and observability vendors are all treating Cursor as a platform to build on. That is a meaningful signal about where the center of gravity in developer tooling is moving.

Cursor's plan structure has already been oriented around agent access. Pro at $20/month gives access to frontier models. Pro Plus at $60/month and Ultra at $200/month unlock deeper cloud agent and code review agent capabilities including Bugbot, Cursor's automated code review agent. Projects capability will almost certainly be tiered toward the higher plans as it exits beta. For an engineering team of ten, even the full Ultra tier across all seats is $2,000/month. If Projects can absorb the equivalent of one sprint's worth of migration or CI maintenance work per month, that math resolves quickly in favor of adoption. The more honest budget question is not the seat cost; it is the cost of the governance infrastructure you need to run agents safely at scale: secrets management, audit logging, policy enforcement on AI-generated PRs. Plan for $500 to $2,000 per developer per month in total tooling cost (seats plus security and observability tooling) when evaluating the full picture. The productivity upside exceeds that threshold by a wide margin for most teams, but you need to budget for the full stack, not just the Cursor seats.

Projects is in beta and rolling out broadly. The teams that run disciplined pilots in the next sixty days will have a meaningful head start on the teams that wait for the case studies. Here is a practical sequence:

Projects accelerates a shift that was already underway: individual feature teams getting smaller and more capable, while engineering organizations as a whole expand to cover more product surface area. A team of three engineers running well-configured Projects coordinators across five parallel workstreams is not a diminished team. It is an elite unit with reach that a team of fifteen could not match six months ago. This is the structural bet that the SpaceX acquisition is underwriting. Not fewer engineers in the world, but a different kind of engineer: someone who thinks in terms of workstream orchestration, agent governance, and context architecture rather than story points and sprint velocity. The engineers who develop fluency with Projects now are not making themselves redundant; they are making themselves significantly more expensive to hire. For engineering leaders, the talent implication runs in parallel with the tooling adoption question. Finding engineers who can operate effectively in an agent-orchestrated workflow is already harder than finding engineers who can write good code. That gap will widen over the next twelve months as Projects and tools like it normalize persistent multi-agent delivery. Your hiring criteria need to evolve in the same timeframe as your tooling stack.

Cursor Projects is the most significant single feature launch in developer tooling in 2026. It is not a productivity increment; it is a delivery model change. Persistent coordinators, shared organizational memory, autonomous CI management, and ecosystem integrations that are already live on day one: this is what an agentic delivery platform looks like when it ships. The teams that adopt it deliberately, with proper governance and a clear pilot strategy, will have a structural advantage that compounds over months. The teams that wait for the market to settle will be catching up to a moving target. Start your pilot. Build your governance layer. The coordinators are already running.