VS Code 1.137 gives each AI agent a container
Many AI agents can work in one editor, each in a container that limits the damage from a bad command.
Microsoft released Visual Studio Code 1.137 on 9 September 2026, and an update came about one week after. The new Agent Host system lets many AI agents work at the same time in the editor. Each agent gets a Dev Container, a closed work area with the tools and parts of the project. The source gives no test results.
How it works
Each agent session gets one work area. A developer can start it, hold it, monitor it and stop it, and the primary editor stays free. For example, an agent can change a module while the developer works on other code.
What else is new
- Access to a session: a person or a different agent can look at a session and change it.
- Codex use: Codex works with Agent Host.
- Automations: tasks that start when a change occurs in the editor or repository.
- Voice mode: voice control of agent sessions.
What is not known
The source gives no speed numbers and no safety tests. Industry analysis tells that the container limits damage, but no test shows it. The source gives no cost.
Insight: Microsoft builds the work area for agents and the test for them
Microsoft makes the editor where agents work. Its tool Scope operates one task with different coding agents. It gives a score to each result, to find where an agent fails with an SDK, a document or a tool. This reads as if Microsoft wants to control the work area and the test, and then it can set the test for agents. If you publish an SDK or tools, test them in Scope.
Sources
Posted