VS Code multi-root AI agents are now an experimental option in version 1.136, released September 2, 2026. Copilot and Claude sessions can work across folders in the editor window’s Chat view. For QA teams keeping application code and test automation in separate folders, this creates a useful evaluation opportunity.
What changed
The official VS Code release notes list separate controls: chat.agentHost.copilotAgent.multiRootEnabled and chat.agentHost.claudeAgent.multiRootEnabled. The feature is experimental and currently limited to the editor window.
There is a key boundary: agent hooks still load from a single workspace folder. When VS Code finds hooks in multiple folders, it asks the user to choose the primary folder. Teams should not assume that opening several projects combines all their hooks.
GitHub’s September 4 weekly release summary also highlights multi-root sessions as experimental. This is a workspace capability update, not a new model announcement.
Why this matters for QA engineers
Our QA takeaway: a shared workspace could help an assistant investigate an API change alongside its integration tests. The hook boundary deserves explicit validation before teams rely on existing test gates during agent work.
For a small pilot, place an application and its test suite in separate folders, then ask the agent to investigate one known regression. Record which files it reads and changes, the working directory used for test commands, and whether the relevant tests actually run.
Next, put distinguishable logging hooks in both folders and verify which primary folder is selected. Compare the observed hook execution with your intended controls. Review the resulting diff and test evidence before accepting the change. These are QATechTools testing recommendations, not vendor claims of improved defect detection.
Verified September 9, 2026; covers the September 2 release and September 4 summary.