Learn Programming with AI as Your Copilot

AI coding assistants can make programming faster, but there's a difference between getting code generated and learning how that code works.

For this experiment, I'm using GitHub Copilot. Other tools such as Cursor, Claude Code, Amazon Q Developer, and Gemini Code Assist follow similar ideas, although their suggestions can differ.

A Small Copilot Experiment

I started with a simple requirement:

function getActiveUsers(users) {
  // Return only users whose status is "active"
}

Copilot suggested:

function getActiveUsers(users) {
  // Return only users whose status is "active"
  return users.filter(user => user.status === "active");
}

Instead of assuming the suggestion was correct, I tested it:

const users = [
  { name: "Alice", status: "active" },
  { name: "Bob", status: "inactive" },
  { name: "Charlie", status: "active" }
];

console.log(getActiveUsers(users));

The result was:

[
  { name: "Alice", status: "active" },
  { name: "Charlie", status: "active" }
]

I also tested an empty array:

console.log(getActiveUsers([]));

Result:

[]

For these test cases, the suggestion worked as expected.

What Did I Learn?

The interesting part wasn't the complexity of the function. It was seeing how much Copilot could infer from a short comment and the function structure.

However, passing simple tests doesn't prove that generated code is suitable for a real application.

A function processing ten users may behave perfectly but raise different questions when processing millions of records. Performance, memory usage, concurrency, and database design can become more important than whether the basic syntax is correct.

What About Bad or Outdated Suggestions?

AI can also suggest deprecated APIs, insecure approaches, or code that doesn't match the version of a framework you're using.

For beginners, the safest approach is don't blindly trust the suggestion.

Check the official documentation, especially for:

• security-sensitive code;

• database queries;

• authentication;

• external APIs;

• framework-specific features;

• code using unfamiliar or deprecated-looking methods.

An AI saying "this is secure" isn't proof that it is secure.

What If the AI Explains Its Own Code Incorrectly?

This can happen too.

If an AI explanation doesn't make sense, run the code and break it into smaller pieces.

For example:

users.filter(user => user.status === "active");

can be understood by examining filter() and the condition separately.

Experimenting with the code yourself is often more useful than asking the AI to explain the same answer repeatedly.

Should Beginners Turn AI Off?

Not permanently, but occasionally.

If you can't write a basic function, loop, condition, or array operation without autocomplete, turn it off and practice those fundamentals.

AI should make you faster, not become the reason you can program.

A Simple Workflow

My recommended workflow is:

1. Try the problem yourself.

2. Ask AI for a suggestion.

3. Read and understand the generated code.

4. Test it with real and edge-case data.

5. Check documentation when necessary.

6. Modify the code yourself.

The most important lesson is simple:

AI-generated code is a proposal, not proof.

The AI can generate the code, but the developer still has to understand it, test it, and decide whether it actually solves the problem.