Visual Studio Code 1.145 reaches Stable with remote agents, cron automations, and a Blobby game
Visual Studio Code 1.145 hit the Stable channel today. The update leans hard into a direction the editor has been moving toward since late 2025: away from a single chat box and toward a persistent, portable "agent host" that keeps AI coding sessions running in the background, crossing windows and even machines.
The headline payoff here is continuity. With the Agent Host and its open Agent Host Protocol (AHP) now maturing, sessions survive folder closes, window closes, and machine changes. You can schedule tasks, mix harnesses in one session, and dial chat verbosity up or down. Keep in mind that Microsoft frames the release around five features: remote agents, cron automations, verbosity controls, file exclusions, and, oddly enough, a Blobby jump game.
It's arguably the most "agent era" release VS Code has shipped to Stable yet. Whether that's a win depends on how much you want your editor doing while you're not looking.
Remote agents that feel local
Two Preview features extend the remote agent host connection support VS Code added around 1.125. First, the integrated browser in the Agents window now loads pages through a remote host. If your agent starts a dev server on a remote machine, you can open it right there in the Agents window to preview, validate, and give feedback; mirroring what you already get with remote workspaces. A screenshot in Microsoft's notes shows the URL bar reading "Connected via kyles-mbp," which is a nice touch if you work across several machines.
Then there are remote canvases. Canvases are the interactive, agent-backed visual workspaces you use to iterate on output. Until now they were pinned to local hosts. With this change you can run them against a remote agent host too, so you never have to leave the Agents window.
Both sit behind experimental settings. Microsoft is asking you to file issues in the VS Code GitHub repository as you test. Head here to see the full release notes.
Control: cron, verbosity, and mixed harnesses
The Agents section is the heart of 1.145. Nested sessions can now use a different harness than their parent. Previously every nested conversation inherited its parent's harness, so getting a "second opinion" meant starting a fresh session and losing the grouping. Now you can ask Claude to review a plan that a Copilot chat wrote, or hand Codex the same task, side by side. Each chat keeps its own model, agent, and permission levels.
Automations let you schedule agent tasks on a set cadence instead of firing them off one at a time. The old built-in presets covered the basics. This release adds five-field cron expressions for when those aren't precise enough. Just know that cron only works for local automations, not the cloud ones. Custom agents from your selected workspace are also usable here.
Chat verbosity is another knob Microsoft added. You can pick between Conversation, Verbose, or Compact modes from the Chat view's More Actions menu. The modes change display only. They don't touch context or the work actually performed, so switching won't reset a session.
It's a rather opinionated change if you've ever wanted to mute the tool-call sprawl. For those who like watching it all happen, Verbose expands every step without an internal scroll limit on tool groups.
Enterprise hardening and the fine print
On the business side, several things finally behave the way admins would expect. The enterprise-managed allowedMcpServers list now takes precedence over registry-only access, so organizations can approve servers without publishing them to their registry. An empty list blocks everything, an absent list keeps the old registry rule, and deniedMcpServers still wins either way.
Organization-required plugins can no longer be uninstalled. Before, the Customizations views offered a failing Uninstall button that flipped to "Retry Uninstall," which also failed. Now it's just disabled, with a hover and keyboard-focus tooltip explaining the plugin is managed by the organization.
There's also better signage when sandboxing can't run. If an org requires sandboxing but the machine can't, a banner now explains the specific cause, like a missing dependency, instead of silently blocking terminal commands. Accounts on cloud sessions get nudged to run agent tools in the cloud instead.
The Local harness gets trimmed
Not everything is new. Microsoft pulled several tools out of the Local harness, mostly around automated workspace creation. #installExtension, #runCommand, and #newWorkspace (alongside the /new slash command) are gone. To create a project now, you open an empty folder and describe it in an agent session. The Generate New Workspace link on the Welcome page went with them, as did the dedicated project-setup skills.
Extension authors should remove references to those tools from custom agents and prompt files, along with the useProjectTemplates and newWorkspaceCreation settings. Notebook creation behavior is unchanged.
Two fixes are worth a mention. Organization-disabled agents no longer trap people in read-only chats; new chats now fall back to the default harness, keeping your unsent prompt. And a failed MCP policy refresh can no longer wipe out a "registry only" restriction until the next successful refresh.
And then there's Blobby. Type /vscode-pet to summon VS Code's in-app pet, or hit Play Jump Game from its context menu on the new-session screen. Hold space to charge, arrow keys to aim. It saves your high score on the device. Not cheap entertainment, but it's a reminder that this editor still has a sense of humor.
Head here to download the new release.
