The VS Code Live stream for the v1.103 release showcased new features including enhanced GitHub Copilot integration, improved Multi-Component Programming tools, and terminal reliability upgrades, with demonstrations from the VS Code and GitHub teams. Additionally, custom AI chat modes like “Beast Mode” and the AI Toolkit’s model playground were introduced to optimize developer workflows, alongside invitations to community events and a GitHub Copilot AMA.
The VS Code Live stream for the v1.103 release kicked off with Olivia, a developer advocate on the VS Code team, welcoming viewers from around the world and introducing the monthly release live stream format. The stream featured multiple guests from the VS Code and GitHub Copilot teams who showcased the latest features and improvements. Olivia emphasized the importance of the open-source community and contributors in shaping VS Code, encouraging viewers to explore the detailed release notes for a comprehensive understanding of all updates.
Bridget, a product manager on the VS Code team, demonstrated the recent enhancements to the GitHub Copilot coding agent integration within VS Code. She explained how the coding agent allows users to delegate coding tasks to Copilot asynchronously, enabling developers to stay focused on other work while Copilot drafts pull requests in the background. Bridget showcased new UI elements like the cloud button for delegating tasks, the chat sessions view for managing multiple agent sessions, and seamless integration with GitHub pull requests, allowing users to review and iterate on AI-generated code directly within VS Code.
Tyler and Connor from the VS Code team presented updates to the Multi-Component Programming (MCP) tools experience, highlighting a redesigned tool picker with a quick tree UI that organizes tools more intuitively and supports sticky scrolling and icons. They also introduced a new system for managing the tool limit imposed by AI models by grouping tools into virtual sets, optimizing tool usage without overwhelming the model. Additionally, they discussed quality-of-life improvements such as automatic MCP server updates and the ability to open chat sessions in separate windows, enhancing user workflow and multitasking capabilities.
Daniel, focusing on the integrated terminal, shared significant reliability improvements made by moving the terminal tool from the extension API to the core VS Code repo. He detailed the technical challenges of shell integration and command execution strategies that underpin terminal interactions with Copilot, resulting in reduced errors and better performance. Daniel also previewed upcoming features like running terminal commands inline within the chat interface and explained the auto-approve functionality for terminal commands, which streamlines workflows by reducing confirmation prompts while maintaining security.
Burke introduced “Beast Mode,” a custom chat mode designed to improve the behavior of the GPT-4.1 model by encouraging it to be more agentic and iterative, reducing the tendency to stall or seek constant user input. He demonstrated how custom chat modes and system prompts can tailor AI behavior to better fit developer workflows, emphasizing the importance of context and prompt engineering. April then showcased the AI Toolkit extension’s model playground, which allows side-by-side comparison of different AI models like Claude and GPT-5, helping users select the best model for their specific coding tasks. The stream concluded with invitations to community events and an upcoming GitHub Copilot AMA, encouraging viewers to engage with the VS Code and Copilot teams directly.