Cloud Agents and Cursor Harness Improvements
02:00 · August 19, 2026 · Cursor Blog

Summary
Cursor has updated its cloud agents and underlying harness to support more autonomous, always-on operation. The changes allow agents to respond to external triggers, maintain objectives across extended sessions, and complete software delivery tasks with reduced manual oversight at each iteration.
Agents can now subscribe to event sources such as pull requests and Slack threads. When a subscribed PR is created, the agent automatically monitors it, addresses CI failures, and responds to automated comments until the review process concludes. In Slack, users can direct the agent to check a thread at a later time and continue working on feedback as it arrives. These subscription capabilities currently apply only to cloud-based agents.
Custom modes let users pin any existing skill directly in the chat interface so the agent remains focused on that capability throughout a session. Subagents have been given dedicated virtual machines, each provisioned with an isolated copy of the project and a clean context. This separation allows multiple subagents to test changes or pursue independent fixes without interfering with one another or the parent agent.
A new /goal command provides a persistent objective that the agent pursues until completion, which can be combined with custom modes or recurring loops. While an agent is executing, follow-up messages are now queued and applied at the next tool call rather than interrupting the current action, preserving workflow continuity.
Why it matters
This update is highly relevant for product teams and builders as it introduces autonomous AI agents into the software development lifecycle, automating PR management, CI/CD fixes, and testing. Dutch AI practitioners can leverage these tools to significantly accelerate development, though they should evaluate the data privacy implications of cloud-based subagents.







