How to Start Using GitHub Copilot Agent Mode

GitHub Copilot is GitHub’s AI pair-programmer suite: inline suggestions, chat, and agent-style workflows that can plan edits, change multiple files, and run tools under your supervision. Product overview: github.com/features/copilot. In VS Code, chat and agent surfaces are documented under Use chat in VS Code and related agent guides.

“Agent mode” (and the broader Agents experience in current VS Code) means you hand Copilot a task—not only a question. It can propose a plan, edit the workspace, and iterate while you stay in review. For after-work Python, that is useful when you have 30–45 minutes and one clear outcome.

Who this is for

People who already live in VS Code or GitHub and want agent help on a personal practice repo after hours. You need a Copilot plan that includes chat/agent features for your account (plans and limits change—check GitHub’s Copilot pages for what your seat includes). Do not point the agent at confidential work trees for a first experiment.

Tool shopping for evenings? See Top 10 AI agent tools for after-work Python.

Steps to start tonight

1. Get Copilot on a supported editor

Install current VS Code from code.visualstudio.com, then install the GitHub Copilot extensions and sign in with the GitHub account that has Copilot enabled. Confirm from github.com/features/copilot that your plan is active. Keep VS Code updated—agent chat features ship with recent VS Code builds.

2. Open a small practice folder

evening-python/copilot-lab/
  scores.py

# scores.py
scores = [88, 72, 95, 60]

def average(xs):
    return sum(xs) / len(xs)

print(average(scores))

Open only this folder as the workspace.

3. Open Chat and choose an agent-capable mode

Open the Chat view (chat icon, or the Chat: Open Chat command; on many builds Ctrl+Alt+I / Ctrl+Cmd+I). In the mode selector, pick Agent (or the agent harness your build exposes) rather than Ask-only chat. If Agent is missing, update VS Code and Copilot Chat, then reload—GitHub/VS Code have rolled agent chat broadly for signed-in Copilot users on current builds.

4. Run one bounded task and review diffs

Prompt example:

Refactor average to return None when xs is empty, add type hints, and keep a single print in __main__-style guard. Do not add packages.

Watch file edits. Use the Changes / diff UI. Accept only what matches the ask. Then run:

python scores.py

5. Prefer permissions you understand

Agent sessions may ask to run terminal commands. For evening practice, approve narrowly (run this script) and deny broad install/network actions you did not request. Start a new chat when the topic changes so context stays clean.

Tiny exercise (15–20 minutes)

Ask Agent mode to add passing_count(xs, threshold=60) -> int and print both average and passing count. If it rewrites the whole file style, reject and say “minimal diff only.”

Pitfalls

• Ask mode vs Agent mode. Ask answers; Agent edits. Wrong mode feels “broken.”
• Huge workspaces. Opening Documents/ makes the agent noisy and risky.
• Auto-approving everything. After-work fatigue + broad permissions is how surprise pip install happens.
• Plan/limit confusion. Features and quotas differ by Copilot plan—verify on GitHub rather than assuming.
• Skipping local runs. Always execute the script yourself before you trust the summary.

Next step

Agent mode is strongest when you can specify data structures, edge cases, and tests in plain English. Build that vocabulary with SoloQuant’s Intro to Python Programming—short after-work lessons, not a weekend bootcamp.