Cursor has introduced a new round of upgrades for its Cloud Agents, adding subscriptions, extending the /goal command, and giving sub-agents their own virtual machines. The update lets cloud agents keep track of pull requests, Slack threads, and scheduled tasks, then resume work automatically when something changes.
Under the new subscriptions feature, a Cursor Agent can attach itself to an event source such as a PR, a Slack thread, or a timed task schedule. Once activity appears, the agent wakes up and continues from where it left off. According to the article, a cloud agent can automatically subscribe to the PR it opened and keep working until that PR is merged. If CI breaks, it attempts a fix. If a bot leaves a comment, it can reply on its own.
Slack is part of the same workflow. Users can tell the agent to come back in an hour and continue after feedback arrives. For now, though, subscriptions are only available to cloud agents. Local agents still need a user prompt before they do anything.
/goal moves beyond CLI testing
The most watched addition in this release is /goal. In a fresh conversation, a user can enter a command such as “/goal fix all flaky tests and make CI green” and leave the agent to keep working until the objective is actually completed.
The piece describes that shift as a move from one-off tasks to outcome-based execution. Cursor’s own CLI changelog had already mentioned the feature, including an active/paused status bar, continued operation across idle and unattended sessions, and support for pausing with Ctrl+C. The official wording there described it as “rolling out and gated,” meaning it was still being gradually released.
According to the article, the Aug. 19 update marks the point where /goal moved from an experimental command-line feature into PR- and Slack-connected workflows.
Sub-agents can now run in separate VMs
The third major change is that sub-agents can each run on their own virtual machine. Every sub-agent gets an independent copy of the project, a clean context window, and its own cloud environment.
Cursor’s stated use case is straightforward: sub-agents can validate changes made by a primary agent in a fresh environment, or multiple sub-agents can tackle different problems at the same time without colliding with one another.
The update also includes two smaller changes. One is Custom modes, which allows any skill to be pinned as a persistent mode. The other is Steering. Instead of interrupting an agent immediately while it is working, a follow-up message now takes effect the next time the agent calls a tool.
Cursor says cloud agents created more than 35% of merged internal PRs
Cursor also published an internal metric in its blog. More than 35% of pull requests that were merged internally were autonomously created by cloud agents. That figure stood at 30% in February 2026, and it was 0 eighteen months earlier.
Auto pricing switches to a per-model basis on Aug. 24
Alongside the product update, Cursor said Auto pricing will change on Aug. 24.
- Included usage across all plans will increase.
- Auto will switch to per-model pricing, with limits raised at the same time.
- Model routing logic will remain unchanged.
- Ultra will include SuperGrok Heavy.
- The changes will take effect in the current billing cycle, and users do not need to take any action.
The article notes that Auto previously used a flat rate per million tokens regardless of which model a request was routed to. Under per-model pricing, requests routed to more expensive models will be billed at the higher model rate.
That change makes the cost of running long-lived cloud agents more visible and more granular. The references cited in the piece include Cursor’s post on X and a post by Da7_Tech on X. The article itself was republished from the WeChat account New Intelligence, with authorship credited to ASI启示录.