The stable VS Code 1.146 release, dated November 11, 2026, deprecates the Local harness and announces its removal in February 2027. It remains usable during the transition but is functionally frozen. The Copilot harness has already been the default for new sessions since version 1.143. The news therefore starts a bounded transition period, rather than abruptly stopping all existing agent conversations.
What is actually changing
A harness is the software that organizes an agent's context, tool calls, permissions and work progression. The official transition guide distinguishes this environment from the selected AI model and from the location where code runs. Choosing another model does not automatically change the harness, while selecting the Copilot harness does not, by itself, move files to the cloud.
The documentation also warns about compatibility: customizations do not all transfer automatically. Hooks have no automatic conversion, and MCP server configuration and authentication, including editor-stored secrets, need verification. This does not say that every user will encounter a problem. It means a previously working workflow does not certify the replacement configuration.
Turning a deadline into a verifiable transition
Our editorial analysis is that a completed task is the most useful unit for this migration. A team might choose a small change in a test repository with a known expected result. It should compare the resulting file, the tools used, access boundaries and approval behavior. This is a proposed verification approach, not an experiment conducted by SIExpo.
Suppose an agent must read internal documentation and then run a test. Those steps may depend on different integrations. Working documentation access does not establish that the terminal has the correct isolation. A successful test does not establish that a subsequent operation outside the working directory will be refused. A positive result and an appropriate denial are separate parts of accepting an agent environment.
The same applies to customizations. An instruction file may be found while another integration is missing. A transition record is therefore more useful when it connects a specific setting to observed behavior. A general note saying the migration passed does not tell the next developer which behavior was checked and which remains an assumption. When a problem appears, preserve the exact example and its conditions without putting secrets in a public report.
Experiments have their own status
The release also includes an experimental Inbox for organizing agent sessions, disabled by default. That status is separate from the stability of the editor itself. A team need not enable every new feature to complete its Local transition. Combining a migration with many optional changes makes it harder to identify which change caused an observed result.
The practical conclusion is to describe the current environment, verify a small representative workflow and then expand the scope gradually. February provides time, but not a promise that an unsupported integration will migrate itself. SIExpo has not changed settings or performed a migration on a user's installation; this article analyzes the officially announced transition and the distinctions needed to assess it.
Sources: Microsoft — Visual Studio Code 1.146, VS Code — move from Local to Copilot harness.
The cover is a conceptual AI illustration, not a documentary photograph.
Editorial verification: 2026-11-13T07:35:06.745176+00:00.
