The latest updates for Cursor are not just about improving the AI coding experience; they represent a fundamental shift in how we can deploy and manage AI agents. With the introduction of self-hosted machines and event subscriptions, agents are moving from being interactive partners to autonomous systems that operate securely within your own infrastructure. This changes the calculus for teams concerned with security and for anyone building automated software delivery pipelines.
run agents on your metal
For many engineering organizations, the primary blocker for adopting powerful cloud-based AI tooling is security. Sending source code, environment variables, and build artifacts to a third-party service is a non-starter. Cursor's new self-hosted machines directly address this.
You can now configure cloud agents to execute on dynamically scheduled pools of machines inside your own network. You manage the infrastructure, and Cursor handles the agent's planning and orchestration. This means your codebase, build outputs, and secrets all remain on internal machines. It's a familiar model for anyone who has used self-hosted runners for CI/CD systems like GitHub Actions, and it provides the control necessary for enterprise adoption.
This gives teams more control over where agents execute and what infrastructure they use, a critical step for integrating agents into trusted, production-level workflows.
agents that subscribe to work
A significant evolution in this release is the concept of "Subscriptions." An agent can now subscribe to an event source, like a conversation or a pull request, and activate when something happens. This shifts the interaction model from command-and-response to a persistent, goal-seeking state.
The most practical example is an agent that automatically subscribes to a PR it creates. It can then drive that PR to completion by fixing CI failures and responding to bot comments without manual intervention. This is a move toward agents that own workflows, not just tasks.
Here’s a conceptual look at what a configuration for such an agent might look like:
# conceptual agent_config.yml
agent:
name: pr-shepherd-agent
trigger:
on: pull_request.opened
filter:
repo: "my-org/my-critical-service"
execution:
runner_pool: "self-hosted-secure-builds"
goal: "Get this PR to a 'passed' CI status and approved by the 'ci-bot'."
permissions:
- code:read
- code:write
- comments:write
- actions:rerun
This configuration defines an agent that wakes up for new pull requests, runs on secure internal hardware, and has a clear, long-running objective. It's a powerful pattern for genuine automation.
a more integrated system
Alongside these major features, other updates focus on reducing friction. Cloud agents now start significantly faster thanks to "Builds," which are pre-built copies of your development environment. Agents boot into a ready state instead of setting up from scratch each time.
Additionally, you can now sync your own GitHub repos directly into the workspace, where they sit alongside repos hosted by Cursor. This makes GitHub the source of truth and ensures the agent is always working with the latest code, further embedding the tool into existing developer ecosystems.
These changes aren't just incremental improvements to an IDE. They are foundational pieces for a future where AI agents are persistent members of the engineering team, capable of running securely and acting autonomously. For builders, this is the pattern to watch; it's how AI moves from assisting with code to taking ownership of the delivery lifecycle.
Top comments (0)