Reviewing AI Code

Time: ~4 minutes | Difficulty: Beginner

What You'll Learn

• What to look for when reviewing AI code
• Red flags that signal problems
• How to build understanding while reviewing

The Big Idea

Never commit code you don't understand. AI code needs review just like human code — sometimes more, because AI can confidently produce incorrect solutions.

The Review Checklist

1. Does It Do What You Asked?

Read through and verify:

• Does it implement your requirements?
• Did AI add things you didn't ask for?
• Did it miss something you specified?

Common issue: AI adds "helpful" extras that complicate things.

2. Do You Understand It?

For each section of code, can you explain:

• What it does?
• Why it's there?

If not, ask:

Prompt: "Can you explain what lines 15-20 do in this code?
Specifically, I don't understand the filter function."

3. Does It Handle Errors?

Check for:

• What happens if the network fails?
• What if the user enters bad data?
• What if required data is missing?

Red flag: No error handling at all.

4. Are There Security Issues?

Watch for:

• Passwords or secrets in code
• User input used directly without validation
• Sensitive data logged to console

More on this in AI Safety Rules.

Red Flags to Watch For

🚩 Overly Complex Solutions

AI says: Here's a 200-line function with 5 nested callbacks...

You should think: Is there a simpler way? Ask:

"This seems complex. Is there a simpler approach?"

🚩 Unfamiliar Libraries

AI says: We'll use mega-form-validator library...

You should think: Do I need another library? Ask:

"Can we do this without adding a new library?"

🚩 Deprecated Patterns

AI says: componentWillMount() in React...

You should think: Is this modern best practice? AI's knowledge has a cutoff date.

🚩 Copy-Paste Errors

AI says: Component name doesn't match filename, or uses wrong variable names.

You should think: Did AI properly adapt to my context?

🚩 No Types (in TypeScript)

AI says: any type everywhere...

You should think: This defeats the purpose of TypeScript.

The "Explain It Back" Technique

Before committing, explain the code to yourself (or your mentor):

"This code:
1. Gets the user email from the form
2. Validates it's a proper email format
3. If valid, calls the API to register
4. If invalid, shows an error message
5. Handles API errors by showing a toast notification"

If you can't do this, you don't understand it enough.

Levels of Review

Quick Review (for small changes)

• Scan the code
• Verify it matches what you asked
• Run and test

Thorough Review (for important features)

• Read line by line
• Look up unfamiliar functions
• Consider edge cases
• Check error handling

Deep Review (for security-critical code)

• Verify authentication logic
• Check data validation
• Review what data goes where
• Consider attack vectors

What to Do When You Don't Understand

Option 1: Ask for Explanation

"Explain this code line by line, assuming I'm a beginner"

Option 2: Ask for Simpler Version

"Can you rewrite this in a simpler way that a beginner could understand?"

Option 3: Ask for Comments

"Add comments explaining what each section does"

Option 4: Ask Your Mentor

Some things are better explained by a human in context.

Building Understanding Over Time

Each time you review code:

1. Note patterns you see repeatedly
2. Look up one thing you don't understand
3. Try to predict what AI will suggest next time

This is how you learn — not by just accepting code, but by understanding it.

Check Your Understanding

• [ ] I should never commit code I don't understand
• [ ] I know the review checklist
• [ ] I recognize common red flags
• [ ] I know how to ask for explanations
• [ ] Review depth matches code importance

Next Up

Now let's learn how to debug effectively with AI help.

Continue: Debugging with AI →