共有:

Coding · Cursor

Cursor 3.7 ships long-horizon cloud agents

Long-running refactors used to lock the local session, leaving you idle until the agent finished. Adds /in-cloud, /babysit, and environment snapshots in the Agents Window — work runs in the cloud while you keep editing locally, landing in Cursor 3.7.

AI Navigate Editorial·2026.06.19·6 min read

Local Machine Agent running... ⏳ Blocked Before 3.7 Cloud Agent /in-cloud running ☁ Local — free Your edits continue After 3.7

The local-lock problem

Handing a large refactor or migration to an AI agent effectively froze the local Cursor session until it finished. While the agent ran, you couldn't focus on other work, creating real productivity waste.

On teams with large codebases, agent task completion waits of 30 minutes to several hours were common, and the developer idle cost added up quickly.

Three new capabilities in 3.7

Adds /in-cloud, /babysit, and environment snapshots in the Agents Window — work runs in the cloud while you keep editing locally.

The /in-cloud command offloads agent processing to the cloud, freeing local resources. /babysit monitors an in-progress agent task and notifies you when intervention is needed. Environment snapshots let long-running tasks resume from a checkpoint if they fail.

Who benefits and how

Heavy migrations can now run in the background while you ship other fixes. Weekly users may not notice.

Engineers who regularly run large-scale refactors or extended data migrations benefit most. Tasks like migrating between microservices, mass file renames, or full test-suite rewrites — things that used to be pushed to off-hours — can now run during the workday.

Users who only generate short scripts or run agent tasks that complete in minutes will feel little difference from local processing.

Source: Cursor official