Vibeleaderboard
← All Intel
Intel / post

Cursor cloud agents hold goals, watch PRs, run subagents on VMs

Source
x.com
Date
Cursor@cursor_ai

We're continuing to improve cloud agents in Cursor. They pick up work from events, hold a goal until it's met, and stay on course through long sessions.

Context

Cursor's post compresses three features from its August 19 changelog into one line. Picking up work from events means a cloud can subscribe to a source such as a pull request or a Slack thread and wake when something happens there, instead of waiting for a person to prompt it again. Holding a goal refers to a new /goal command, which gives the agent a long-lived objective to keep pursuing until it is done; Cursor's example is starting a chat with the goal of fixing flaky tests and getting the test suite passing.

The two combine in pull request following. According to the changelog, a cloud agent now subscribes automatically to the pull requests it opens and keeps working on them, including fixing failed automated checks and responding to comments left by review bots. In Slack, a user can ask the agent to check back later and keep going until feedback arrives. Cursor says subscriptions are limited to cloud agents for now.

The same release lets an agent hand work to subagents that each run on their own virtual machine, with a separate copy of the project and a clean . Cursor suggests using this to test the parent agent's changes in a fresh environment or to run independent fixes in parallel without them colliding. For staying on course, the changelog's closest features are custom modes, which pin a in the chat so the agent stays focused on it, and steering, where a follow-up message waits for the agent's next tool call instead of cutting it off mid-action. The changelog gives no pricing or plan details.

Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
  • agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
More from Cursor
Recommended reads
Comments

Checking sign-in…

Loading comments…