What is worth reading about AI this week.

Self-hosted coding-agent workers improve execution control only when the cloud control plane and data flows remain explicit

Source: Cursor

Cursor has introduced Self-Hosted Machines, allowing Cloud Agents to execute tool calls on dynamically scheduled workers inside a team's network while Cursor continues to run the agent loop, planning, and inference. Workers keep a long-lived outbound HTTPS connection to Cursor; they can be organised into shared pools that scale with queued work and serve multiple repositories. Cursor notes that tool output may include code and that agent transcripts may be processed and stored in its cloud.

Why this matters: Treat this as a split-responsibility design, not a simple self-hosting claim. Before connecting a worker, document the repositories, services, credentials, network routes, operating systems, browsers, and hardware it can reach; isolate pools by trust boundary; enforce least-privilege runner identities and egress rules; and decide what tool output or transcript content can leave the environment. Also retain an approval, logging, revoke, and incident-response path for every worker pool.