VS Code 1.144: Cloud Sessions, Copilot Fusion, Image Generation, and More (A Friendly Guide)

You start an agent task, close your laptop, and the session stops with it. We've all been there. 😅

VS Code 1.144 (released October 28, 2026) is built around fixing that kind of friction. It makes agent work keep going across environments, helps you pick the right model, and gives you more control over what agents change.

So what actually changed, and what's worth trying first? Let's walk through it.

VS Code 1.144 is the latest monthly-style update of Visual Studio Code, and it's mostly about agents. An agent is an AI helper that can work on a task for you: read code, edit files, run tools, and report back.

Here's a simple way to picture it. Think of an agent as a helpful teammate. Earlier, that teammate sat at your desk and left when you left. Now, that teammate can work somewhere else (like in the cloud) and you can check in from anywhere.

Some of these are marked Preview or Experimental, so expect them to keep changing.

Working with agents isn't only about writing code. It's about where the work runs, which model does it, and how much you trust it to do on its own.

VS Code 1.144 gives a practical answer to each one. Fewer interruptions and less guesswork means more time for the actual problem. 🚀

Cloud sessions let you run Copilot tasks from VS Code without keeping your development machine online. Your repository and tools run in a managed environment on GitHub's infrastructure.

Example: You start a task, then head home. Later you check progress on GitHub.com or in the GitHub Mobile app, and continue in VS Code when you're back.

One thing to know: cloud sessions use a separate workspace, so local uncommitted changes aren't automatically copied. They also need a paid Copilot plan and follow your organization's policies.

The new Remote Environments dialog brings GitHub Environments, Dev Tunnels, WSL distributions, and SSH hosts together. You can see each remote's status, connect or disconnect, and choose which remotes show up in the pickers.

Example: You have an SSH host and a WSL distro. Instead of hunting through menus, open Manage Remote Environments at the bottom of the Environment picker and handle both.

Cloud automations run on GitHub infrastructure, on a schedule or in response to repository events, even when your computer is off. You can create one from the Automations view in the Agents window, or ask an agent to set it up.

Example: Trigger an automation when you're assigned an issue with a specific label, such as triaging it automatically. You can limit the allowed tools and add filters for tighter control.

Select Copilot Fusion in the model picker and Copilot chooses the models, reasoning effort, and execution pattern per request. A simple task might use one model. A harder one might have one model draft a solution while others review it.

You pick an effort level: Standard, High, or Max.

Example: Renaming a variable doesn't need a heavy setup. A tricky refactor might. Fusion makes that call so you don't have to.

Model comparisons run the same prompt across models (and harnesses) so you can review results side by side. They're enabled by default for eligible new sessions in the Agents window. This release expands them, so you can now compare from a folderless chat or from a local or empty repository with no commits.

Example: You're unsure which model writes better tests for your code. Run the same prompt on four and compare. Note that model comparisons aren't supported in Dev Containers.

Agents can generate images with GPT Image 2.5 Flare. You can ask for an icon, a chat background, or a website wireframe, or attach an existing image and ask for changes or variations.

It uses AI credits in addition to those used by the conversation model.

Agents can make a lot of tool calls. Approving each one is safe but slow. Giving the agent full freedom can be risky.

Assisted permissions (now generally available for Copilot sessions) use an LLM-based risk check on each tool call. Low-risk calls are approved automatically. Higher-risk calls, or ones where the risk can't be determined, still ask you.

In the Copilot harness, edits made through agent tools land in a working set. You can use inline Keep and Undo controls for single changes, or manage everything from the working set above the chat input.

Example: The agent fixes your bug but also touches a file you didn't expect. Undo that one file and keep the rest. Edits made through terminal commands don't appear in the working set, so keep that in mind.

With Computer Use, the agent can see and control apps on your computer. It works in Copilot harness sessions on your own macOS or Windows machine. The agent asks permission to turn it on, or you can run /computer on (and /computer off to stop).

Copilot connectors are GitHub-managed MCP servers that connect your agent to services without local server setup. The release notes mention Slack, Notion, and Microsoft Learn as examples.

Browse is the new home page for Customizations, with plugins and skills from marketplaces you or your enterprise configured. For Business and Enterprise, an administrator must enable Editor preview features.

If you're wondering how the permission choices differ, here's a simple view:

Assisted permissions require Folder isolation and might be restricted by your organization. If your admin disables it, it stays visible but greyed out, with an explanation.

1. Expecting cloud sessions to see local uncommitted work. They use a separate workspace. Commit or push first.

2. Looking for a feature that isn't showing up. Many items are Preview or Experimental and behind settings. Examples: chat.editor.workingSet.enabled, chat.agentHost.computerUse.enabled, and chat.canvases.enabled. Also, some features depend on your plan or organization policy.

3. Thinking the terminal is covered by the working set. Edits from terminal commands don't appear there. Review those changes another way.

4. Treating model choice as a one-time decision. New models keep arriving. Comparisons and Fusion exist so you can re-check without guessing.

5. Skipping the side notes for extension authors. If you publish extensions, vsce publish --oidc lets you publish from GitHub Actions without storing a PAT as a repository secret. It needs @vscode/vsce 4.0.2 or later. Also, global PATs in Azure DevOps are retired on December 1, 2026, so Azure Pipelines users should move to Microsoft Entra ID with a service connection.

VS Code 1.144 makes agent work feel less tied to one machine and one model. Cloud sessions and automations keep things moving, Copilot Fusion and model comparisons help you choose well, and Assisted permissions plus the working set keep you in control.

My suggestion: don't try everything at once. Pick one or two things, like model comparison and Assisted permissions, and see how they fit your workflow. If you're not on the update yet, use Check for Updates in VS Code. 😊

If this guide helped, you can find more developer posts at hamidrazadev.com. Share it with a teammate who's still approving every tool call by hand. And tell me in the comments: which VS Code 1.144 feature are you trying first? 👀