Summary
Cursor's agent-execution infrastructure now spans managed cloud-agent environments, desktop cloud-subagent setup, a /automate skill for recurring automations, and — as of September 2, 2026 — self-hosted machines that keep tool execution entirely inside a team's own network. The through-line is that Cursor is making remote and self-hosted execution, environment definition, and always-on automation editor-native rather than a separate platform workflow.
What changed
Cursor added and expanded cloud-agent development-environment controls, then added the /automate skill with GitHub and Slack triggers and computer-use support for Cursor Automations. On September 2, 2026 it shipped self-hosted machines: Cloud Agents keep their reasoning loop in Cursor's cloud while executing tool calls on the customer's own infrastructure, via My Machines (a single laptop or VM) or autoscaling Team pools, across AWS Lambda, Coder, Cloudflare, Daytona, Modal, Namespace, Vercel, and E2B, with computer-use support on Linux and Mac.
Why it matters
Environment setup, security policy, multi-agent cloud supervision, and recurring automation are becoming editor-native for teams running long-lived coding agents against real internal stacks. Self-hosted machines close the biggest enterprise gap: source code, build outputs, and secrets never leave the customer's network, making Cursor Cloud Agents adoptable inside security- and compliance-constrained organizations.
Evidence excerpt
Cursor's changelog describes cloud environment setup, cloud subagents in the Agents Window, and /automate support for GitHub and Slack triggers; the September 2, 2026 entry adds self-hosted machines that "keep tool execution entirely in your own network," with My Machines and autoscaling Team pools running on AWS Lambda, Coder, Cloudflare, Daytona, Modal, Namespace, Vercel, and E2B.