VS Code is getting a dedicated agent workspace, but its local AI story is still tied to GitHub Copilot and the cloud. That makes the new Agents window useful, and also revealing.
Microsoft is pushing Visual Studio Code deeper into the agent era with a new Agents window, a preview workspace built less like a code editor and more like a command center for software agents. The interesting part is not just that VS Code now has another surface for managing AI work. It is that even local AI remains wrapped inside the Copilot system.
According to Microsoft's VS Code documentation published on May 13, 2026, the Agents window can be opened from the title-bar button, the Command Palette, the VS Code welcome page, or directly from the terminal with code --agents. Once open, it shows agent sessions across workspaces, lets developers start new sessions, review generated file changes, validate work locally, and move between projects without opening each repository in a separate editor window.
That is a real workflow change. Coding agents are no longer being treated as a side panel next to the editor. They are being given their own room. For developers who already run multiple agent sessions across client work, internal tools, experiments, and production repositories, this could be cleaner than keeping chat panels and terminals scattered everywhere.
The catch is in the requirements. Microsoft says the Agents window requires GitHub authentication so it can access a user's Copilot subscription and sessions. It also shares underlying sessions with the main VS Code window, currently including Copilot CLI, Copilot Cloud, and Claude agent sessions. That makes it convenient if your AI workflow already lives inside Copilot. It is less compelling if your workflow is built around local or independent tools.
The limitation matters because Microsoft's own docs say the Agents window does not yet support local or other third-party agents in that dedicated surface. Those still have to be managed from the main VS Code window. In other words, the agent-first interface is available, but only for the agent types Microsoft has chosen to bring into that surface first.
This distinction sounds small until you look at how developers actually use these tools. A local model is not just a cheaper chatbot. It is often about control. Teams use local inference because they want tighter data boundaries, less dependency on outside services, more predictable costs, or the ability to keep working in restricted environments. If the interface that manages agent work still depends on Copilot identity, Copilot session infrastructure, and a supported agent list, then the local part only goes so far.
Local AI Still Has a Cloud Door
VS Code does support local models through language model provider extensions and bring-your-own-key style configurations, including providers that can connect to locally hosted models. But Microsoft's separate language model documentation is clear on the point developers are arguing about: using a local model currently still requires access to the Copilot service, a GitHub account with a Copilot plan such as Copilot Free, and an online connection.
That creates an awkward message. The model may run on your machine, but the product experience is not offline-first. For a hobbyist experimenting with a local model, that may be a minor annoyance. For a company evaluating whether AI coding can work inside a stricter security environment, it is a bigger architectural question. The word local carries expectations. Developers hear it and think autonomy. Microsoft is using it inside a managed Copilot framework.
There is also a business logic here that should not be ignored. VS Code is one of the most important developer tools in the world, and Microsoft has every reason to make Copilot the organizing layer for agentic development. If developers manage tasks, sessions, file changes, validation, and pull requests through Copilot-aware surfaces, the IDE becomes more than a place to write code. It becomes the control plane for AI software work.
That strategy is powerful because it does not require locking developers out of alternatives outright. VS Code can support Claude sessions, local models, extension-provided tools, MCP servers, and custom agents, while still making the most polished path run through GitHub authentication and Copilot entitlements. The door is open, but the hallway is Microsoft's.
For many developers, the practical question is simple. Does this make daily work faster? If the answer is yes, the Copilot requirement may be accepted as the cost of a better interface. The Agents window's ability to centralize sessions across workspaces is genuinely useful, especially as agent work moves from one-off prompts to longer running tasks that need review and validation.
But the broader market signal is just as important. The next phase of AI coding will not be decided only by model quality. It will be decided by who controls the workspace where agents plan, edit, test, and ship code. Microsoft's new Agents window shows exactly where it wants VS Code to sit in that future. Local AI may be part of the picture, but for now, Copilot still holds the keys.
Also read: Germany is turning AI security procurement into a sovereignty test • Nvidia puts Kimi K2.6 on a faster path to Blackwell inference • Figma is turning AI from a design threat into new revenue
