Mastering GitHub Copilot: How to Use Your AI Coding Assistant Well

Key takeaways

• Copilot suggests code and answers questions, but you remain responsible for correctness and security.
• Good prompts state the goal, inputs, outputs and constraints. Context from open files helps.
• Generate tests too, and run them. Never accept code you cannot explain.
• Check your organisation's rules before using it on private or client code.

What Copilot does

GitHub Copilot is an assistant built into editors and the GitHub website. The main ways to use it:

• Inline completions: suggestions appear as you type or after a comment.
• Chat: ask questions, request explanations, refactors or tests in a side panel.
• Inline edits: select code, describe a change, review the proposed diff.
• Agent-style tasks: in supported tools it can plan and make multi-file changes you then review.

Which features and models you get depends on your plan and tool version. Check the GitHub Copilot documentation for the current plans, features and settings, since they change often.

Setup in VS Code

1. Make sure your GitHub account has Copilot access (individual, organisation or free tier if offered).
2. Install the GitHub Copilot extensions in VS Code and sign in.
3. Open a project and start typing a function name or a comment. Press Tab to accept a suggestion, Esc to dismiss it.
4. Open the chat panel for questions.

Prompting that works

State the goal, the inputs and outputs and any constraints. Compare:

• Weak: # parse the file
• Better: # Read a CSV of orders with columns id, amount, date. Return total amount per month as a dict. Skip rows with a missing amount.

An example in Python. Prompt in a comment:

# Return the n most common words in text, ignoring case and punctuation.

A typical suggestion looks like this:

import re
from collections import Counter

def top_words(text, n):
 words = re.findall(r"[a-z']+", text.lower())
 return Counter(words).most_common(n)

It looks right. Now review it. It handles the basic case, but does it treat "don't" as one word (yes, because of the apostrophe)? Does it handle non-English letters (no, the pattern only matches a-z)? Ask Copilot Chat to write tests, then run them.

def test_top_words():
 assert top_words("Hi hi, hello!", 1) == [("hi", 2)]

The point: suggestions are drafts. The review step catches limits like the Unicode one.

Habits that make it useful

• Keep relevant files open so it has context.
• Ask for an explanation of code you inherit, then verify it.
• Ask for tests and edge cases before accepting a function.
• Use it for boilerplate such as data classes, simple scripts and regular expressions, then check them.
• Break big tasks into small prompts.

Where it fails

• It can produce code that looks correct but is subtly wrong, uses outdated APIs or invents functions that do not exist.
• It can suggest insecure patterns, such as string-built SQL or weak crypto. Review with security in mind; see the OWASP Cheat Sheet Series.
• It does not know your business rules unless you tell it.
• Relying on it too early can slow learning. Beginners should write core exercises themselves first.

Security, privacy and licensing

• Do not paste secrets, keys or customer data into prompts or code.
• Check your employer's or client's policy before using it on private repositories.
• Review the settings on suggestions that match public code and the data-use options for your plan in the documentation.
• Run linters, tests and security scanners on all generated code, like any other code.

Next steps

Read about AI extensions for VS Code and open-source coding assistants with Granite models to compare options. To build the fundamentals the assistant cannot replace, see the Python programming course or the AI application development course.

Related reading

• Which Code Editor Is Best for Python, Java, C++, JavaScript, PHP & More in 2026?
• [2026] Top 50+ Cloud Development Tools Interview Questions and Answers
• Cyber Defense Reinvented | How Large Language Models Are Transforming Threat Intelligence with CYLENS