The latest Visual Studio Code update introduces enhanced multi-agent support, allowing users to manage and run multiple coding agents—local, cloud, background, or third-party—simultaneously for streamlined development workflows. New features like the Agent Sessions welcome page, session management sidebar, and parallel subagents enable efficient task delegation, context isolation, and seamless integration with tools like GitHub Copilot and external AI providers.
The latest release of Visual Studio Code introduces a revamped experience for managing multiple coding agents, whether they are local, cloud-based, background, or third-party providers like Claude or Codex. Users can now leverage several agents simultaneously to handle various coding tasks, streamlining their workflow. A new Agent Sessions welcome experience appears at startup, offering an overview of all agent sessions and allowing users to quickly jump into existing sessions or start new ones. The Agent Session sidebar, accessible via the chat icon in the Command Center, provides ongoing access to session management.
To optimize the startup experience, users are encouraged to set the Workbench Startup Editor to the Agent Sessions Welcome page, though this feature is still experimental. When initiating an agent session, developers can choose from different agent types: local agents for interactive work, background agents for isolated changes in a Git worktree, cloud agents for cloud-based changes resulting in pull requests, and even integrate external providers like Claude or Codex as local agents. This flexibility allows for a tailored approach to coding tasks, such as using a local plan agent to research and suggest improvements for app accessibility while simultaneously working with other agents.
The workflow is further enhanced by the ability to track and manage agents through a session counter at the top of the editor, filter sessions by status, and maximize chat input while agents continue their work. For example, a background agent can be started to asynchronously add new features to an app, such as a page for likes and dislikes, using the Copilot CLI in a Git worktree. Meanwhile, a cloud agent can be tasked with updating documentation, like the README file, and automatically creating a pull request through the Copilot coding agent, all managed under a single GitHub Copilot subscription.
Once agents complete their tasks, users can review the results, iterate on plans, or implement changes directly. For instance, after a local plan agent finishes suggesting accessibility improvements, the user can choose to implement those changes. Similarly, background agent-generated changes can be reviewed and applied to the workspace, followed by a Copilot code review and automated commit message generation, enabling a seamless workflow from prompt to push.
A standout new feature is the ability to run parallel subagents. These are context-isolated agents that operate independently from the main chat session, allowing for cleaner context management and faster results. For example, a user can launch subagents to analyze error handling across different layers of a project—front end, back end, and database—simultaneously. Each subagent works on its specific task, and their results are consolidated in the main chat. The VS Code team continues to innovate and encourages user feedback to further enhance the agent user experience, aiming to make VS Code the ultimate platform for agent-driven development.