Copilot Agent Mode
Agent mode is GitHub Copilot’s local, synchronous autonomous mode — it edits your files, runs commands, and loops until the task is done, all inside your IDE while you watch.
In regular Copilot chat, you ask → Copilot answers → you apply. In agent mode, you give a goal → Copilot plans → acts → checks results → repeats. It can touch multiple files, run the terminal, and recover from its own errors. You stay in control — you can interrupt at any point.
Agent mode is NOT the Copilot Coding Agent
⚠️ Agent mode is NOT the Copilot Coding Agent. Despite both having “agent” in the name, these are separate products. Agent mode runs in your IDE right now, synchronously. The Copilot Coding Agent runs in the cloud, asynchronously, and creates a pull request. If you came here looking for the cloud-based async agent, that’s the other page.
What agent mode can do
- Edit multiple files in a single task
- Run terminal commands (build, test, lint)
- Read error output and self-correct
- Iterate through a multi-step plan
- Use MCP tools (if configured)
What agent mode cannot do
⚠️ Agent mode has real limits — know them before you use it:
- It does NOT work asynchronously — you must be present
- It does NOT create pull requests automatically
- It does NOT have access to GitHub issues or PR context
- It does NOT run on GitHub’s infrastructure — it runs on your machine
How to use it
- Open Copilot Chat in your IDE
- Switch the mode selector to Agent
- Describe the goal (not the steps) — e.g., “Add input validation to the registration form and write tests for it”
💡 Tip: Give it a goal, not instructions. Agent mode is better at “fix all the TypeScript errors in this project” than “edit line 42 and then edit line 87.”
Which model does agent mode use?
Agent mode uses whichever model you’ve selected in Copilot. You can switch models mid-conversation to trade off speed, cost, and reasoning depth. See Copilot Models for the full list of available models and their trade-offs.
Dive deeper
- For the full picture of agents and orchestration, see What is an Agent? and Models vs Agents
- For the cloud-based alternative, see Copilot Coding Agent
Related Concepts
- What is an Agent? — The concept behind agent mode
- Copilot Coding Agent — The cloud-based, asynchronous alternative to agent mode
📍 Layer 2 — The Agent (IDE-local) · Back to the AICA Onion
What’s next?
- Prompting Agent Mode — learn how to write effective prompts for agent mode
- Copilot Coding Agent — the cloud-based, asynchronous alternative that creates pull requests