Enable GitHub Copilot Agent Mode in VS Code in 5 Minutes

Problem: Your Copilot Only Suggests Code, Doesn't Execute It

You're using GitHub Copilot for autocomplete but still manually running tests, fixing errors, and editing multiple files yourself. Agent mode changes that—it autonomously plans, edits, tests, and iterates until your task is complete.

You'll learn:

• How to enable agent mode in VS Code (1.99+)
• The difference between Ask, Edit, and Agent modes
• When to use agent mode vs traditional Copilot
• How to control autonomous tool execution

Time: 5 min | Level: Beginner

Why This Matters

GitHub Copilot launched agent mode in February 2025, shifting from code suggestions to autonomous execution. As of January 2026, there are 4.7 million paid Copilot users, with agent mode becoming the default workflow for complex tasks.

What agent mode does:

• Creates multi-step plans before coding
• Edits files across your entire workspace
• Runs Terminal commands (build, test, lint)
• Monitors output and self-corrects errors
• Iterates until the task succeeds

Common symptoms without agent mode:

• Manually applying suggested code changes
• Copy-pasting fixes across multiple files
• Running tests yourself after each edit
• Context-switching between editor and terminal

Prerequisites

Before enabling agent mode:

Required:

• VS Code version 1.99 or later
• Active GitHub Copilot subscription (Free, Pro, Pro+, Business, or Enterprise)
• GitHub Copilot extension installed

Check your version:

code --version

Expected: 1.99.0 or higher

If you're on an older version:

# Update VS Code
# macOS: Code > Check for Updates
# Windows/Linux: Help > Check for Updates

Solution

Step 1: Verify Copilot Extension

Open VS Code and check that GitHub Copilot is active.

# Open Command Palette
Cmd+Shift+P (macOS)
Ctrl+Shift+P (Windows/Linux)

# Type: GitHub Copilot: Sign In

Expected: "Signed in as [your-github-username]" in the status bar

If it fails:

• Not installed: Go to Extensions (Cmd+Shift+X), search "GitHub Copilot", install
• Not signed in: Run the sign-in command again and authenticate via browser

Step 2: Enable Agent Mode

Agent mode is controlled via VS Code settings.

Method 1: Settings UI

1. Open Settings (Cmd+, on macOS, Ctrl+, on Windows/Linux)
2. Search for chat.agent.enabled
3. Check the box to enable

Method 2: Settings JSON

{
  "chat.agent.enabled": true,
  "github.copilot.chat.agent.runTasks": true,
  "github.copilot.chat.agent.autoFix": true
}

Why these settings:

• chat.agent.enabled - Activates agent mode in the chat picker
• runTasks - Allows agent to run workspace tasks (build/test)
• autoFix - Enables automatic error detection and correction

Expected: Agent mode now appears as an option in Copilot Chat

Step 3: Access Agent Mode

Open the Copilot Chat view and select Agent mode.

# Open Copilot Chat
Ctrl+Cmd+I (macOS)
Ctrl+Alt+I (Windows/Linux)

# Click the mode dropdown (top of chat panel)
# Select "Agent" (options: Ask, Edit, Agent)

Mode differences:

• Ask: Conversational Q&A, no code edits
• Edit: Single-file inline changes with preview
• Agent: Autonomous multi-file execution

Step 4: Test Agent Mode

Run a simple multi-file task to verify agent mode works.

Prompt:

Create a new TypeScript utility function called formatDate in src/utils/date.ts, 
then write a test file for it in src/utils/__tests__/date.test.ts, 
and run the tests to verify it works.

What the agent does:

1. Creates a plan with 4-5 steps
2. Asks for confirmation before executing
3. Creates src/utils/date.ts with the function
4. Creates src/utils/__tests__/date.test.ts with tests
5. Runs npm test (or your test command)
6. Monitors test output
7. Fixes failures if any occur

Expected output:

✓ Plan created (3 steps)
  1. Create date utility file
  2. Create test file  
  3. Run tests

✓ Created src/utils/date.ts
✓ Created src/utils/__tests__/date.test.ts
Running tests...
✓ Tests passed (1 suite, 3 tests)

Task complete.

If it fails:

• "Agent mode not available": Verify chat.agent.enabled: true in settings
• No terminal execution: Enable runTasks setting
• Agent asks for manual approval: This is normal—review the plan before accepting

Step 5: Configure Tool Permissions

Agent mode requests permission before running tools. Configure this based on your trust level.

Permission options:

• Always ask (default) - Manual approval for each tool
• Allow for this session - Auto-approve until VS Code restarts
• Allow for this solution - Auto-approve for current workspace
• Remember my choice - Auto-approve forever

Recommended for beginners: Keep "Always ask" enabled until you're comfortable

To reset permissions:

// settings.json
{
  "github.copilot.chat.agent.requireConfirmation": true
}

Verification

Test agent mode with a realistic scenario:

Prompt:

Refactor the UserService class to use async/await instead of promises, 
update all tests, and verify they still pass.

You should see:

1. Agent analyzes UserService.ts
2. Creates a refactoring plan
3. Applies changes to the class
4. Updates test files automatically
5. Runs tests
6. Reports success or iterates on failures

Success indicators:

• Multiple files edited without your intervention
• Terminal commands executed automatically
• Agent recovers from build/test failures
• Final state: all tests passing

Advanced Configuration

Enable MCP Servers

Agent mode supports Model Context Protocol (MCP) for external tools.

{
  "github.copilot.chat.agent.enableMCP": true
}

Available MCP servers:

• GitHub MCP (default): Repository search, issue management
• Database MCP: Query databases from chat
• Cloud MCP: AWS/GCP/Azure integrations

To add an MCP server:

1. Install MCP extension from VS Code marketplace
2. Configure in .copilot/mcp.json
3. Restart VS Code

Customize Models

Agent mode works with multiple LLM providers.

Change model:

# In chat, type:
/model

# Select from:
# - Claude Sonnet 4.5 (best for complex tasks)
# - GPT-4o (balanced performance)
# - Gemini 2.0 Flash (fastest)

Note: Claude Sonnet 4.5 achieves 56% on SWE-bench Verified—best for autonomous coding tasks

Limit Agent Actions

Prevent agent from editing specific files.

Create .copilot/ignore:

# Exclude from agent edits
*.config.js
package.json
.env

Or use workspace settings:

{
  "github.copilot.chat.agent.excludeFiles": [
    "**/*.config.js",
    "**/package.json"
  ]
}

What You Learned

• Agent mode enables autonomous multi-file coding with self-healing
• Enable via chat.agent.enabled: true in VS Code settings
• Agent creates plans, edits code, runs tests, and iterates on failures
• Use Ask mode for questions, Edit for single files, Agent for complex tasks

Limitations:

• Agent mode consumes premium requests (300/month on Pro, 1,000 on Enterprise)
• Best on smaller repos (<10k files); large codebases may timeout
• Requires manual review—don't blindly accept all changes
• MCP servers not available on Copilot Free plan

When NOT to use agent mode:

• Simple one-line code completions (use autocomplete)
• Refactoring core architecture without human oversight
• Tasks requiring deep domain knowledge agent can't infer
• Projects with complex build systems agent struggles to understand

Troubleshooting

Agent Mode Not Appearing

Issue: Agent option missing from mode dropdown

Solution:

1. Verify VS Code version >= 1.99
2. Check chat.agent.enabled: true in settings
3. Restart VS Code
4. Update GitHub Copilot extension to latest version

Agent Stops Mid-Task

Issue: Agent creates plan but doesn't execute

Causes:

• Tool permissions denied
• Workspace folder not set
• Terminal profile misconfigured

Fix:

{
  "chat.tools.terminal.terminalProfile.osx": "zsh",
  "chat.tools.terminal.terminalProfile.windows": "PowerShell",
  "chat.tools.terminal.terminalProfile.linux": "bash"
}

Enterprise Policy Blocking

Issue: Agent mode grayed out in organization

Solution:

• Contact your GitHub admin to enable agent mode in organization policies
• Go to github.com/organizations/[org]/settings/copilot/policies
• Enable "Agent mode" under Copilot features
• May require Copilot Business or Enterprise plan

Tested on VS Code 1.99.4, GitHub Copilot v0.24+, macOS Sonoma & Windows 11