Cursor is expanding its cloud agents with support for event-driven work, persistent goals and longer-running sessions. The company announced the changes on August 19, describing agents that can pick up tasks from external events and continue working until an objective is complete.
The updates move Cursor’s agents beyond single prompt-and-response tasks. Agents can now monitor PRs, follow Slack threads and run scheduled tasks, according to the company. Cloud agents also automatically subscribe to the PRs they create and continue working on them through completion.
Cursor has added a /goal command for long-lived objectives. A command such as /goal Build a software factory gives an agent an ongoing target rather than a one-off instruction. The agent is intended to keep pursuing that objective through a long session until it is marked complete.
That persistence could reduce the need for repeated prompts, although it also introduces a familiar risk with autonomous coding tools: an agent may continue pursuing a flawed assumption. One commenter questioned what checkpoints would be available to redirect an agent before it spends hours working toward the wrong result. Cursor’s announcement does not detail new review or interruption controls beyond its updated steering behavior.
More control over agent behavior
Cursor is also turning skills into pinned Custom Modes. A skill selected from the / menu can remain active in chat, functioning as an “always-on” instruction set. On Mac, the shortcut is Option+Enter; on Windows, it is Alt+Enter. The interface also offers a “Use as Mode” option.
The company has changed steering so that a follow-up waits for the agent’s next tool call instead of interrupting it mid-action. Users can send a follow-up immediately or press Enter twice. That should allow an agent to finish its current operation before incorporating new direction.
The update also adds isolated virtual machines for subagents. Each subagent can work from a separate copy of the project, allowing it to test a parent agent’s changes in a clean environment or handle an independent fix at the same time. Cursor presented the feature as a way to run parallel work without putting every agent in the same project environment.
Source: Cursor on X
