Cursor Cloud-Agents: Coden, even when the laptop is shut down

18 June 2026

5 min. read

Cursor now lets its coding agents keep working in the cloud, even after the laptop has long been closed. Switching to “Build in Cloud” in Plan mode hands execution over to a cloud agent that runs in an isolated Ubuntu environment, allowing developers to continue working locally or simply shut down their machines. The agent finalizes the pull request, delivers demos and screenshots for review, and can be toggled back and forth between the cloud and local environment with a single click. While this sounds like pure convenience, it actually shifts the fundamental question of where software is built-and consequently, where it needs to be secured.

Key Takeaways

Coding agents move to the server: Cursor runs background agents in isolated cloud VMs that operate independently of the local editor and continue working even when the laptop is closed.
The benefits are real: Parallel tasks, a prepped pull request, and seamless handoffs between cloud and local environments lighten the load for developers and speed up workflows.
The bill comes later: Server-side agents consume cloud compute and, depending on the configuration, access repositories and secrets. Cost and build security must be addressed before rollout, not after.

What Cursor Actually Changed

The core shift is a change in execution. Cursor’s standard agent works inline within the editor on local files, tackling one task at a time. Cloud agents, by contrast, run asynchronously in their own isolated Ubuntu environments. Following an expansion in early 2026, each agent now receives a full development environment complete with a desktop and browser, meaning it can interact with UIs and visually validate results.

In practice, this means developers can kick off a task with “Build in Cloud” in Plan mode and then shut down their machines. The agent continues iterating server-side, prepares the pull request for merge, and never blocks a local session. If you want to take over the results, you can pull them back into your local environment with a click and send them back to the cloud if needed. According to the Cursor documentation, cloud agents generate demos and screenshots along the way, making it easy to verify their work without diving into the code.

Why this matters for DevOps teams

The obvious gain is concurrency. Multiple agents can work on different tasks simultaneously without blocking the developer’s local machine. This shifts the human role from typing to oversight: defining tasks, reviewing results, and approving pull requests.

For DevOps teams in DACH, the more interesting point is the proximity to the pipeline. An agent that runs all the way to a finished pull request deeply intervenes in the development flow. That demands clear guardrails: Which repositories may it access, which actions may it trigger, and who ultimately approves? Without these rules, a productivity tool can quickly turn into uncontrolled automation. The question of control over autonomous agents is no longer just about Cursor-it’s now on every IT leader’s mind.

The bottleneck also shifts. When agents deliver code in minutes, the review becomes the new choke point. A team that used to handle a handful of pull requests per day suddenly faces multiples that all look plausible. This is where real productivity gains are decided: if review capacity isn’t scaled accordingly, faster delivery only piles up risk. Practical criteria are essential-define which changes an agent may prepare autonomously and which always require human review, especially anything touching dependencies, migrations, or security logic.

What sets cloud agents apart

Isolated VM each agent gets its own Ubuntu environment, independent of the editor.
Laptop off execution runs server-side; the pull request is prepared remotely.
Handoff switch between cloud and local environments with one click, complete with demos and screenshots for verification.

The cost question remains

Server-side agents incur costs. Depending on plan and quota, every cloud VM iterating in the background adds up in consumed compute time-and the total grows when a team keeps several agents running continuously. Those who fail to track inference and compute costs may face the same shock that has derailed many AI projects once the AI bill blew past the cloud budget.

The honest calculation weighs saved developer hours against ongoing cloud charges. In many cases it pays off because a parallel agent replaces expensive idle time. Still, run the numbers before rolling the tool out widely-otherwise the cloud invoice arrives only after the budget is already overspent.

The build environment becomes an attack surface

The second blind spot is security. A cloud agent preparing a pull request operates on the repository and, depending on setup, may access build secrets or even deployment rights. Those credentials now reside in an environment the developer no longer controls directly. An isolated VM helps, yet it doesn’t replace proper role-based access control.

Treat the agent like another build runner: minimal permissions, short-lived tokens, clear separation between read and deploy rights, and an audit log of everything it touched. Introducing server-side agents without this foundation enlarges the attack surface right where code is built and delivered.

Add the supply-chain angle. An agent that adds or updates packages makes dependency decisions that later land in production. Without reviewing those suggestions, responsibility for the software supply chain quietly shifts to a model. Teams already relying on signed dependencies and reproducible builds should apply the same scrutiny to whatever a cloud agent contributes.

Frequently Asked Questions

What are cloud agents in Cursor?

Asynchronous agents running in isolated Ubuntu cloud environments, independent of your local editor. They continue working on tasks even when your machine is shut down and prepare pull requests.

Can I really shut my laptop?

Yes. In Plan mode, “Build in Cloud” hands off execution to a cloud agent that keeps running server-side. You can later pull the results back into your local environment with a single click.

What benefits do DevOps teams gain?

Parallelism and a ready-made pull request. Multiple agents work simultaneously without blocking your local session, shifting human effort to defining and approving tasks.

What does it cost to run?

Each running cloud VM consumes compute time depending on your plan and quota. Those costs should be weighed against the developer hours saved before rolling the tool out widely.

What security concerns arise?

Depending on configuration, the agent operates on the repository and may access build secrets or deployment rights. Treat it like a build runner: minimal permissions, short-lived tokens, and an audit log of its actions.