GitHub announced on November 16, 2026 that Visual Studio Code 1.146 deprecates the Local harness, the layer that coordinates model, context and tools in GitHub Copilot agent sessions inside the editor. Removal is planned for February 2027. Its replacement, the Microsoft Copilot harness, is built into VS Code and has been the default for new sessions since version 1.143.
What goes away and what replaces it
A harness is the program that sits between the model and the machine: it decides what context reaches the model, which tools the agent may call and when it asks for confirmation. The Copilot harness is built on the Copilot SDK and uses the same agent runtime as Copilot CLI and the GitHub Copilot app, which lets a session continue between the editor and the Agents window.
According to the announcement, the switch does not move files to the cloud: agent commands keep running on the developer's machine. The subscription plan does not change either, Copilot Free included, although individual features depend on the plan and on organization policy.
There is a three-month transition before removal. During that period the Local harness stays available where organization policy allows it, and new improvements go only to the Copilot harness.
Who needs to act
Anyone who starts new sessions with the default settings has been on the Copilot harness since 1.143. The work falls on two groups.
The first is teams that still run sessions on the Local harness and built customizations on top of it. GitHub warns that switching harnesses does not automatically convert every customization, so instructions, tools and agent workflows built for the Local harness need to be tested again.
The second is VS Code administrators at companies that keep the vscode.temporaryDefaultToLocalHarness exception. That managed setting is temporary and will be removed along with the Local harness. Keeping it enabled does not extend anything past February.
What to do before February
For each developer, the action is to start a new session, choose Copilot under Session Target, and then review customizations using the VS Code migration guide.
For administrators, security is the part that needs the most care. The announcement says the Copilot harness can enforce managed runtime permissions and sandbox restrictions that the Local harness cannot, but it also says that selecting the Copilot harness does not turn on the sandbox. Those rules have to be configured separately. An organization that treats the migration as a change of default ends up with the same level of control as before, only on a different runtime.
Then there are assisted permissions: a risk assessment approves some agent calls and sends the rest to the person for confirmation. That cuts interruptions, but it hands part of the decision to a classifier. It is worth deciding beforehand which actions should never go through without human confirmation, the same reasoning as in agent autonomy starts with limiting the damage.
In my reading, the deadline is tight for companies that validate tooling changes in quarterly cycles. Starting the review in December leaves room to report a blocking workflow in the VS Code repository, as GitHub suggests.
What is still unclear
The announcement says February 2027, with no exact day and no VS Code version number for the removal. It also does not list which kinds of customization stop working in the conversion. Copilot connectors and canvases, cited as advantages of the new harness, are still in preview, and Copilot Fusion tiers vary by plan. Anyone who depends on one of these should follow the notes for upcoming releases before treating the migration as done.
If your company is weighing something similar, reach me on LinkedIn or WhatsApp.