TLDR: Microsoft has released Visual Studio Code (VS Code) version 1.104, focusing on significant advancements in AI model selection and agent security. Key updates include an ‘Auto model selection’ feature for optimal performance in chat, a new API for extensions to contribute language models, and enhanced security measures requiring confirmation for sensitive file edits by AI agents.
Microsoft has rolled out Visual Studio Code (VS Code) version 1.104, the August 2025 update, introducing a suite of new functionalities primarily centered around artificial intelligence (AI) model management and agent security. This release, announced on September 11, 2025, aims to provide developers with more control, flexibility, and safety when working with AI-powered tools within the popular code editor.
Advanced AI Model Selection and Contribution
A standout feature in VS Code 1.104 is the ‘Auto model selection’ in chat. This intelligent capability allows VS Code to automatically choose the most suitable AI model to ensure optimal performance and help users avoid hitting rate limits. The system currently selects from a range of powerful models including Claude Sonnet 4, GPT-5, GPT-5 mini, GPT-4.1, and Gemini Pro 2.5. Users can view the actively selected model by hovering over the ‘Auto’ option in the model picker. Furthermore, paid GitHub Copilot users benefit from a variable model multiplier and a 10% request discount.
Beyond automatic selection, VS Code 1.104 finalizes an API that empowers extensions to contribute their own language models. This opens up the ecosystem for broader AI integration, with examples already seen in extensions like AI Toolkit for VS Code, Cerebras Inference, and Hugging Face. Microsoft has provided comprehensive guidance for extension authors in the Language Model Chat Provider documentation.
Enhanced Agent Security Measures
Security has been a critical focus in this update, particularly concerning AI agent interactions. The agent mode now incorporates a crucial safeguard: it prompts users for confirmation before making modifications to sensitive files. This protection extends to common system folders, dotfiles, and any locations outside the user’s workspace. Administrators can fine-tune these patterns using the `chat.tools.edits.autoApprove` setting. Additionally, improved controls for terminal auto-approval and refined prompts further bolster the guardrails around commands executed by AI agents.
Productivity Improvements for Developers
Several productivity enhancements have also been integrated into VS Code 1.104. The introduction of `AGENTS.md` allows developers to include a file at the workspace root, which is automatically utilized as context for chat requests. This feature is enabled by default and can be managed via the `chat.useAgentsMdFile` setting. Other improvements include the ability to reference custom chat modes from prompt files and configure which prompts appear in the Chat welcome view. The ‘Changed files’ user experience has also been refined, now collapsing by default, automatically removing accepted items, and displaying counts and line deltas for a cleaner workflow with agent suggestions.
Also Read:
- Microsoft Azure Unveils ‘Agent Factory’ Initiative, Championing Open Standards for AI Agent Interoperability
- Microsoft Unveils Role-Specific Copilots for Enhanced Productivity Across Business Functions
This update underscores Microsoft’s ongoing commitment to integrating advanced AI capabilities into developer tools while prioritizing performance, security, and user experience.
