The video, presented by Josh, showcases the new agent collaboration features in Visual Studio Code (VS Code) 1.109, demonstrating how four types of agents—local, cloud-based, background, and plan mode—can streamline coding tasks and enhance workflow using a real-world project example. It highlights the flexibility of running multiple agent sessions, seamless handoff between agent types, and integrated session management, all aimed at boosting developer productivity within VS Code.
The video, presented by Josh, focuses on the latest advancements in agent collaboration within Visual Studio Code (VS Code), particularly following the 1.109 release, which introduced numerous agent-related features. Josh begins by highlighting the extensive updates and documentation around agents, explaining that agents in VS Code are now categorized into four main types. The session aims to demonstrate practical use cases for each agent type, using a Bluetooth game project called “Tank Game” as a real-world example to showcase how agents can assist in adding new features and improving workflow.
Josh first discusses the local agent, which operates directly on the user’s machine using VS Code’s built-in agent harness. This agent can edit code, answer questions, and operate in different modes such as “agent mode” for direct code editing and “plan mode” for brainstorming and planning new features. A notable improvement is the ability to run multiple local agent sessions concurrently, allowing users to multitask without risk of file conflicts, especially when using plan mode. Josh demonstrates how the agent can analyze the codebase, suggest new features, and even implement them automatically.
The video then covers third-party integrations, specifically with cloud-based agents like Claude (by Anthropic) and Codeex, which can be selected within VS Code. These agents provide similar experiences to the local agent but run in the cloud, allowing tasks to continue even when the user’s device is offline. Josh illustrates how cloud agents can be used to prototype features, create pull requests, and manage code changes remotely, all integrated seamlessly within the VS Code interface. This is particularly useful for tasks that require less hands-on involvement or need to be executed while the user is away from their computer.
Next, Josh introduces the background agent, which leverages the Copilot CLI and Git worktrees to create isolated environments for each task. This allows multiple background sessions to run in parallel without interfering with each other, making it ideal for handling several independent features or experiments simultaneously. The session management panel in VS Code helps users keep track of all ongoing, completed, and pending agent sessions, providing filters to manage and review work efficiently.
Finally, Josh highlights the flexibility of agent handoff, which enables users to start a session in one mode (such as local or plan mode) and seamlessly transfer it to another agent type, like background or cloud, as the project evolves. This interoperability ensures that developers can choose the most productive workflow for their needs, combining the strengths of different agent harnesses from VS Code, Anthropic, and OpenAI. The video concludes by emphasizing the goal of empowering developers with versatile, integrated AI tools that enhance productivity and streamline collaborative coding within VS Code.