GitHub Copilot & Coding Assistants

The intern who shipped in three days

Priya joined a startup as a junior developer. Her first task: build an API integration with Stripe to handle subscription billing. She'd never worked with Stripe before. Without AI tools, this would mean two weeks of reading documentation, writing boilerplate, debugging edge cases, and asking senior developers for code reviews.

Instead, Priya opened Cursor (an AI-powered code editor). She typed a comment: // Create a Stripe checkout session for a monthly subscription plan. Cursor generated 30 lines of working code — the correct API call, error handling, and response formatting. She reviewed it, adjusted the plan IDs, and tested. It worked.

Over three days, she built the entire billing system: checkout, webhooks, subscription management, and invoice retrieval. Her tech lead reviewed the code and found it cleaner than what most mid-level developers would write. Not because Priya was a prodigy — because she knew how to guide an AI coding assistant and verify its output.

By the end of this module, you'll understand how AI coding assistants work under the hood, know the key differences between Copilot, Cursor, and Codeium, and be able to spot bugs in AI-generated code.

How AI coding assistants actually work

AI coding assistants are large language models trained specifically on code. They read what you've written so far — your file, your project, your comments — and predict what comes next. Think of them as autocomplete on steroids: instead of suggesting the next word, they suggest the next 5-50 lines.

They work at three levels:

1. Line completions — finish the current line as you type
2. Multi-line suggestions — generate entire functions or blocks from a comment or function signature
3. Chat/agent mode — have a conversation about your code, ask for refactoring, debugging, or explanations

The tools compared

There Are No Dumb Questions

"Do I need to be a programmer to use these tools?"

Yes and no. These tools accelerate existing developers — they don't replace the need to understand code. You still need to read, review, and debug what the AI generates. However, non-programmers can use tools like Replit AI or Cursor to build simple projects with heavy AI assistance. Think of it as the difference between driving a car and being a mechanic.

"Will AI replace software developers?"

No — but it's changing the job. Developers who use AI tools write code 30-55% faster (depending on the task and study). The work is shifting from "typing code" to "designing systems, reviewing AI output, and handling the hard problems AI can't solve." Junior developers who learn AI tools have an advantage over seniors who refuse them.

Setting up GitHub Copilot (5 minutes)

What these tools are actually good at

Not all coding tasks benefit equally from AI assistance.

Where AI coding assistants excel:

• Writing boilerplate (CRUD operations, API routes, database queries)
• Generating unit tests from existing code
• Writing documentation and comments
• Regular expressions and complex formulas
• Translating code between languages
• Implementing well-known patterns (auth flows, pagination, form validation)

Where they struggle:

• Novel algorithm design
• System architecture decisions
• Understanding complex business logic unique to your company
• Security-critical code that needs careful review
• Performance optimization for specific hardware
• Code that requires deep domain expertise (medical, financial, scientific)

Prompting techniques for coding assistants

1. Comment-driven development

This is the coding equivalent of the structured prompting from the Claude module — specificity in, quality out. Write a detailed comment, then let the AI generate the code.

# Function that takes a list of transactions and returns:
# - Total revenue (sum of amounts where type is "sale")
# - Total refunds (sum of amounts where type is "refund")
# - Net revenue (total - refunds)
# - Average transaction value
# Returns a dictionary with these four keys

The AI will generate the complete function matching your specification.

2. Provide examples in context

If you want code in a specific style, write one example first. The AI will follow the pattern.

// Example: GET /api/users - Returns all users
app.get('/api/users', async (req, res) => {
  const users = await db.users.findMany();
  res.json({ data: users });
});

// GET /api/products - Returns all products
// (AI generates this following the pattern above)

3. Use chat mode for complex tasks

Instead of inline completions, open the chat panel and describe what you need:

"Refactor this authentication middleware to:
1. Check for both JWT and API key auth
2. Return 401 with a specific error message for each failure case
3. Add rate limiting of 100 requests per minute per user
4. Log all auth failures to the audit table"

When NOT to trust AI-generated code