AI has transformed how developers write and debug code. Tools like GitHub Copilot, ChatGPT, and Claude can generate functions, explain errors, suggest fixes, and even architect entire applications. For beginners, AI can be an always-available tutor. For experienced developers, it is a powerful accelerator.

But AI-generated code comes with risks that are different from the risks in AI-generated text. A factual error in an essay is embarrassing. A bug in production code can be catastrophic. This lesson covers how to use AI for coding effectively and safely.

---

How AI Writes Code

AI models generate code using the same next-token prediction mechanism as they use for text. During training, they were exposed to billions of lines of code from open-source repositories, documentation, Stack Overflow, tutorials, and technical books.

This means AI is particularly good at:

• Common patterns: Code that appears frequently in training data (sorting algorithms, API calls, database queries) is generated reliably
• Well-documented languages and frameworks: Python, JavaScript, TypeScript, React, Django, Rails — anything popular with extensive documentation
• Boilerplate code: Repetitive setup code, configuration files, standard CRUD operations

And less reliable at:

• Novel or unusual architectures: Custom patterns that are not well-represented in training data
• Business logic: Complex, domain-specific rules that require understanding the specific problem
• Security-sensitive code: Authentication, encryption, input validation — where subtle bugs can have serious consequences
• Performance-critical code: Optimisation often requires understanding of specific hardware and runtime characteristics

---

Use Case 1: Explaining Error Messages

One of the most valuable AI coding use cases is error message explanation. Many error messages are cryptic, especially for beginners.

Good Prompt Pattern

"I am getting this error in my Python code. Explain what it means in simple terms and suggest how to fix it:

TypeError: cannot unpack non-iterable NoneType object

Here is the relevant code: ```python def get_user_data(user_id):

• ... database query ...*

• sometimes returns None*

name, email = get_user_data(123) ```"

Including the actual error message AND the relevant code gives the model everything it needs. Without the code, the model can only give generic explanations.

---

Use Case 2: Describing Bugs

When something is not working as expected but you do not have a clear error message, AI can help — but only if you describe the problem clearly.

The Bug Description Formula

1. What you expected to happen: "The function should return a list of unique names"
2. What actually happened: "It returns a list with duplicates"
3. The relevant code: Paste the function and any related code
4. What you have already tried: "I tried using set() but it changed the order"

Example

"This JavaScript function should filter an array to keep only items where the price is under £10, but it is returning an empty array even when there are items under £10. Here is the code:

```javascript function filterCheap(items) {

• return items.filter(item => item.price < '10');* } ```

What is wrong?"

The AI will quickly spot that the price is being compared to the string '10' instead of the number 10 — a classic type comparison bug in JavaScript.

---

Use Case 3: Generating Boilerplate Code

Boilerplate — repetitive setup code that follows standard patterns — is one of AI's strongest coding use cases.

Examples of Good Boilerplate Requests

• "Create a basic Express.js server with routes for GET, POST, PUT, and DELETE on a /users endpoint"
• "Write a Python class that connects to a PostgreSQL database with methods for basic CRUD operations"
• "Generate a React component with a form that has fields for name, email, and message, with basic validation"
• "Write a Dockerfile for a Node.js application that runs on port 3000"

Why This Works Well

Boilerplate code is heavily represented in training data and follows well-established patterns. The model has seen thousands of Express servers, React forms, and Dockerfiles, so it generates highly reliable code for these tasks.

The Important Caveat