6 Game-Changing VS Code AI Features You'll Love

New AI features are being added to VS Code almost every day. In this article, I want to show you five—actually, six, if you stick around for the bonus one at the end—new features that I think you're absolutely going to love. These will change the way you use agent mode and code completions every single day inside of VS Code. So, let's get into it.

1. Simplified MCP Server Management

First and foremost, it's now easier to see what MCP (Model Contributor Protocol) servers you have installed, either inside your project or globally for your user account in VS Code, and also to install new ones.

If you go over to the Extensions view, you can see your installed and recommended extensions. You'll also see a new MCP Servers section here. If you don't have any installed at a user level or inside your project, you can click on "Browse MCP Servers." This will take you to a page where you can browse different available MCP servers.

There are tons available, including a curated list of popular services like GitHub, Docs, Markdown, Hugging Face, Notion, Zapier, and many more.

Let's say you want to install the GitHub MCP server to have it available globally. You would click the install button, which opens VS Code to a page that lets you explore the MCP server, view its manifest, go to its repository, and see setup instructions. It's a lot like installing a regular extension.

Just click Install, and it's ready to go. You can then go into its settings to start the server, view output, and configure it. Starting the server will prompt you to log in to GitHub. Once authenticated, you'll see it in your available tools.

This process automatically configures the MCP server in your user-level settings, so you don't need to manually create an mcp.json file. If you use the command palette (Ctrl+Shift+P or Cmd+Shift+P) and search for "MCP," you'll find an option to "Open User Configuration." This will open the JSON file where the server is configured.

Of course, if you prefer to manage this on a per-project basis, you can simply copy this configuration into a new mcp.json file in your project's .vscode directory.

2. Automated Command Execution with an Allow/Deny List

When using agent mode, you're often adding features, fixing bugs, and performing other tasks that require building the project or running tests. Normally, if you ask the agent to "build this solution," it will identify the necessary command, like dotnet build, and then ask for your permission to run it.

This confirmation step can slow you down. Now, you can specify which commands the agent can run automatically without prompting.

Go to Settings, then navigate to Extensions > GitHub Copilot. Under the Experimental section, you'll find Agent: Command Allowlist and Agent: Command Denylist. Here, you can add any command you trust to run without confirmation. For example, you could add dotnet build, dotnet run, and dotnet test to the allowlist.

With this configured, when you ask the agent to build the project, it will run the command immediately, streamlining your workflow.

3. Uninterrupted Agent Mode with max_requests

Long-running operations in agent mode can sometimes be interrupted. By default, the agent will stop and ask for permission to continue after 20 requests.

You can change this behavior with a new setting. Go to your settings and search for max_requests. This setting, GitHub > Copilot > Max Requests, defaults to 20. You can increase this to a much higher number, like 100 or even 10,000, to let the agent work uninterrupted until the task is complete.

Combining this with the command allowlist creates a much smoother, more autonomous agent experience.

4. Auto-Generated Copilot Instructions

A powerful way to improve the agent's performance is to provide it with clear, detailed instructions about your project. The "Generate Instructions" feature automates this process.

In the Copilot Chat view, click the sparkle icon and select Generate Instructions. This will analyze your entire codebase, including its structure, coding conventions, and best practices. It then creates or updates your Copilot instructions, giving the agent a comprehensive overview of your project.

Every time you interact with the agent, it will use these instructions to provide more accurate and context-aware assistance. As your project evolves, you can regenerate the instructions to keep them up-to-date.

5. Snooze Code Completions for Focused Coding

Code completions (the "ghost text" that appears as you type) are incredibly helpful, but sometimes you need to focus without suggestions popping up.

You can now temporarily disable this feature. Click the Copilot icon in the bottom status bar. A menu will appear, allowing you to configure completions for the current file type or disable them entirely. You'll also see a Snooze option. This lets you pause completions for five-minute intervals.

When snoozed, you'll still get standard IntelliSense, but the AI-powered ghost text will be suppressed, letting you type without interruption. A small "Z" icon will appear on the Copilot status icon to indicate that it's snoozing. Click it again to re-enable completions at any time.

Bonus: Beast Mode for Advanced AI Planning

The official Copilot repository on GitHub contains a wealth of resources, including custom prompts and chat modes that tailor the agent's behavior for specific tasks.

One of the most impressive is "beast mode," created by Burke Holland. This mode enhances the agent's planning and execution capabilities, particularly when using advanced models like GPT-4. It guides the agent through a more rigorous, multi-step process of understanding requirements, planning a solution, and then executing that plan.

To use it, you find the beast-mode.md file in the repository, copy its raw content, and save it as a new chat mode file in your VS Code settings. Once enabled, you can select "beast mode" when starting a new chat session. The agent will then follow a more detailed and structured approach, breaking down complex tasks into smaller, manageable steps and providing updates along the way. This is a game-changer for tackling larger, more complex coding challenges with the AI agent.