Visual Studio Code 1.104 rolls out with auto model selector and AI safety features

If you’re a GitHub Copilot user on an individual plan, there’s good news. Microsoft has added auto model selection to Visual Studio Code’s chat feature in the August 2025 (v1.104) update.

Instead of you manually switching between AI models, the tool now automatically picks the best one. You can choose from Claude Sonnet 4, GPT-5, GPT-5 mini, GPT-4.1, or Gemini Pro 2.5. This eventually helps coders balance performance and avoid rate limits.

If you’re a paid user, you even get a 10% request discount and variable multipliers. Organizations, however, retain control, with the ability to disable access to specific models.

If you’ve been worried about AI agents editing critical files, Microsoft has you covered too. A new chat.tools.edits.autoApprove setting now requires explicit user confirmation before sensitive files, like configs, are touched.

Similarly, the terminal auto-approve flow has been updated with mandatory opt-in, and the global auto-approve setting has been renamed with a clear warning label to prevent accidents. These safeguards directly respond to user concerns and observed risky behavior.

If productivity is your focus, then there’s more good news. A new AGENTS.md file allows you to give instructions to AI agents, while TODO comments now come with quick Code Actions to launch coding sessions. Chat and terminal experiences have been streamlined too, featuring collapsed file lists, multi-session support, and better IntelliSense. Math rendering in chat using KaTeX is now enabled by default.

Besides AI improvements, Windows users can now add colored window borders via the window.border setting. Editors can show tab indexing with workbench.editor.showTabIndex. Python developers benefit from AI-powered hover summaries in Pylance and improved environment variable handling. Finally, support for bower.json IntelliSense has been dropped, given Bower’s deprecation since 2017.