VS Code 1.128 runs multiple Claude sub-agents and listens to global hotkeys

The biggest change in this release centers around the agent workflow. Developers using the Claude agent can now run multiple chats inside a single session, effectively branching out their thought process. Instead of throwing everything into one chaotic dialogue, it is now possible to have one thread building an API endpoint, a second thread writing unit tests for it, and a third exploring completely different implementations, all while sharing the same context.

These parallel conversations keep their own history and selected model, and the entire structure restores automatically when the editor restarts. For quick questions that do not require project files, a new feature allows starting fast chats outside of any workspace, bypassing the usual workspace selection clutter completely.

Furthermore, those who want to watch their AI sub-agents do the heavy lifting can now open read-only subagent chats to monitor the delegation process without risking breaking the main context. On the visual front, GitHub Copilot Vision is now fully public, meaning developers can drag and drop screenshots, error diagrams, or PDF specifications directly into the chat to let the AI analyze them.

Customization fans who bring their own AI keys can now tweak temperature and top_p parameters directly in JSON, while corporate admins get central control over GitHub Copilot telemetry through OpenTelemetry. To make navigating the app easier, a new setting controls where built-in browser tabs open, while power users can now define system-wide global keyboard shortcuts that trigger editor actions even when the application is running in the background.

As code editors morph into multi-layered dashboard managers for autonomous AI laborers, the classic definition of a programmer is rapidly dying. Soon, the primary skill on a resume might just be the ability to referee arguing bots without getting a headache.

Source: Visual Studio Code