VS Code runs AI agents remotely

Visual Studio Code’s May 2026 releases moved AI agents off the local laptop and onto remote machines. In version 1.121, Microsoft added support for connecting the Agents window to a remote machine over SSH or a development tunnel, where VS Code installs or connects to its CLI and starts what it calls an “agent host.” The feature matters because the agent loop is no longer tied to the editor process on a developer’s own device. Microsoft’s documentation says the remote agent host is a long-lived process built on the Copilot SDK, and the company separately documents support for third-party cloud agents inside VS Code through GitHub Copilot. ### How does the remote setup actually work? The Agents window connects to a remote machine using what Microsoft calls the Agent Host Protocol over SSH or a dev tunnel. (code.visualstudio.com) When the connection is made, VS Code automatically installs and starts the VS Code CLI on the remote machine if needed. Microsoft’s 1.121 release notes say the editor can either download and install the CLI over SSH or connect to a running CLI server through a dev tunnel that the user has already started. (code.visualstudio.com) The remote side then launches the agent host, which keeps the agent session running outside the local editor. ### Why did Microsoft pair the feature with security warnings? Microsoft’s documentation says developers using dev tunnels should require authentication, because a tunnel without authentication can expose the machine to anyone with the tunnel URL. (code.visualstudio.com) That warning sits alongside the remote-agent documentation because the feature depends on remote access paths that can outlive a single editing session. (code.visualstudio.com) The company also describes the remote agent host as a long-lived process, which means the security question is not only who can open the editor, but who can reach the machine and its running agent session. Microsoft’s broader guidance on AI agents has emphasized “governance, observability, and security” as the controls needed for wider adoption. ### Is this just a one-off VS Code feature, or part of a broader product push? (code.visualstudio.com) Microsoft has been framing VS Code as a hub for “multi-agent development” since at least February 2026. In a VS Code team blog post, the company said it wanted one place to run agents, manage sessions and choose tools without switching editors. The documentation footprint around 1.121 points in the same direction. (code.visualstudio.com) Microsoft now has separate pages for the Agents window, third-party agents in VS Code, and enterprise configuration for AI features, alongside the monthly release notes. ### What changed after 1.121? Visual Studio Code 1.122, published May 19 and last updated May 21, added another remote capability: agents can now trigger tasks on remote machines. (code.visualstudio.com) That extends the model from running an agent remotely to letting it invoke remote tasks as part of the workflow. The sequence matters. Version 1.121 established the remote agent host and connection model, and version 1.122 added task triggering on the remote machine days later. (code.visualstudio.com) Microsoft’s release archive lists both versions in its 2026 release stream. ### Where does this leave developers and tool vendors? Microsoft’s own AI security messaging says organizations need shared governance, visibility and accountability around agents. (code.visualstudio.com) That language matches the kinds of tooling now appearing around agent execution: access controls, monitoring, and policy enforcement around what remote agents can do and where they run. Microsoft’s next public milestones are already on the calendar. (code.visualstudio.com) The company’s Build conference in San Francisco is scheduled for June 2, 2026, where Microsoft-backed startup and developer announcements are expected to continue, and VS Code’s monthly release cadence means another product update could follow within weeks. (news.microsoft.com 1) (news.microsoft.com 2)