VS Code Agents preview

Software used to mean one developer, one editor window, one branch. Visual Studio Code’s new Agents app flips that into several artificial intelligence coding sessions running at once, each in its own repository worktree, with one screen to watch the changes come in. (code.visualstudio.com) A worktree is a second checkout of the same Git repository, like opening two copies of the same document on different desks so edits do not collide. Microsoft says the preview app can kick off agent sessions across multiple repositories in parallel and keep each one isolated in its own worktree. (code.visualstudio.com) That sounds small until you picture a real day of engineering work. One agent can update a user interface repository, another can fix a service in a backend repository, and a third can write tests, while the human reviewer moves between diffs instead of typing every line. (code.visualstudio.com) Microsoft has been building toward this for months. Visual Studio Code 1.109 in January 2026 added ways to manage multiple agent sessions in parallel, and version 1.110 in March 2026 added more control over longer-running agent tasks inside the editor. (infoworld.com) (code.visualstudio.com) The April 8, 2026 release of Visual Studio Code 1.115 moves that idea into a separate companion app that ships with Visual Studio Code Insiders. Microsoft describes it as built for “agent-native development,” which is its way of saying the software is designed around supervising agents, not just writing code by hand. (code.visualstudio.com) (infoworld.com) The app is not just a launcher. Microsoft says developers can track session progress, review inline diffs, leave feedback for agents, and create pull requests without leaving the app, which turns it into a control room for code generation and review. (code.visualstudio.com) This sits on top of another shift inside Visual Studio Code itself. Microsoft’s agent mode, introduced in preview in February 2025, can already refactor across multiple files, run tests, migrate code, and answer questions about a codebase from inside the editor. (code.visualstudio.com) Put those pieces together and the job starts to look different. The editor becomes the place where an agent works inside one codebase, and the new companion app becomes the place where a human juggles several agent workers across several codebases at once. (code.visualstudio.com 1) (code.visualstudio.com 2) That raises a boring problem that turns expensive very quickly: provenance. In a regulated codebase for healthcare, finance, or government work, a reviewer may need to know which agent changed which file, what prompt triggered it, what tools it used, and whether the output crossed a boundary the architecture was supposed to enforce. (code.visualstudio.com 1) (code.visualstudio.com 2) Microsoft has already been adding guardrails for that world. Recent releases added agent permissions, hooks that run before or after agent actions, and background terminal controls, all of which are ways to limit what an agent can do and to record more of how it did it. (code.visualstudio.com 1) (code.visualstudio.com 2) So the real change is not that Visual Studio Code got one more artificial intelligence feature. It is that Microsoft is turning the editor into a multi-agent workspace where review discipline, repository isolation, and change tracking matter as much as raw speed. (code.visualstudio.com) (code.visualstudio.com)