Agent Mode
Agent mode transforms Copilot from a reactive assistant into an autonomous worker. Instead of responding to one question at a time, it plans a sequence of actions, edits multiple files, runs terminal commands, reads error output, and iterates until the task is done — all without you approving each individual step. You describe the outcome; Copilot figures out the steps.
How Agent Mode Works
Write a natural language description of what you want done — 'add pagination to the user list API endpoint'
Copilot reads the relevant files, understands the codebase structure, and determines which files need to change and in what order
Copilot makes changes across multiple files — models, controllers, tests, configuration — in sequence
Runs terminal commands (tests, builds, linters) and reads their output to check if the changes work correctly
If a test fails or a build errors, Copilot reads the error, adjusts the code, and tries again — without asking you each time
What It Can Do
Multi-file edits
Change an API endpoint, update its tests, adjust the client-side call, and update relevant types — all in one task.
Terminal commands
Run `npm test`, `pytest`, `cargo build`, or any shell command and read the output to verify success.
Scaffold features
Create a new feature from scratch — new files, routes, components, tests — based on a description of what it should do.
Refactoring
Rename a function across all usages, extract a class, migrate from one library to another, apply a new pattern project-wide.
Bug investigation
Given a failing test or error, trace through the code to find the root cause and fix it, running tests after to confirm.
Dependency updates
Update a package, scan for breaking changes, apply fixes across all affected files, verify tests pass.
How to Use It in VS Code
- Open the Copilot Chat panel (Ctrl+Shift+P → “GitHub Copilot: Open Chat”)
- Click the mode selector at the top of the chat — switch from Ask to Agent
- Type your task description — be specific about what outcome you want, not how to achieve it
- Copilot will show a plan and begin making edits — you can watch the changes in real time in the editor
- You can pause, undo changes, or redirect mid-task if the approach is wrong
Agent mode is available on Copilot Pro, Business, and Enterprise. Not available on the free tier.
Good Tasks for Agent Mode
- Well-defined, bounded features: “add email validation to the signup form with error messages”
- Repetitive pattern application: “add logging to every API endpoint in this service”
- Test coverage: “write unit tests for all functions in auth.ts that currently have no tests”
- Migrations: “migrate from Moment.js to date-fns across the whole project”
Limitations & What to Watch For
- Scope creep — agent mode may make broader changes than intended if the task description is vague. Review diffs carefully before accepting.
- Wrong assumptions — if the codebase has unusual patterns, Copilot may make changes that technically work but don't follow your conventions.
- Terminal permissions — you will be asked to approve terminal command execution before Copilot runs commands; you can review and block specific commands.
- Doesn't understand external state — Copilot only sees what's in the filesystem. It can't know about database state, running services, or environment-specific configurations.
Checklist: Do You Understand This?
- Agent mode is autonomous — plans tasks, edits files, runs commands, iterates on errors without per-step approval
- Available in VS Code on Pro/Business/Enterprise plans (not free tier)
- Enable via the mode selector in the Chat panel — switch from “Ask” to “Agent”
- Best tasks: bounded features, repetitive patterns, test writing, library migrations
- Watch for: scope creep, non-standard convention violations, terminal command execution (requires approval)