It's the 2025 version of "Stack Overflow Copy-Paste." But instead of copying a single function, junior developers are now generating entire files. As a manager, how do you spot it? And more importantly, how do you fix it?
The Goal Isn't to Ban AI
Let's be clear: banning ChatGPT is like banning Google. It's a losing battle, and it hurts productivity. The goal is to ensure your team is using these tools as assistants, not replacements for critical thinking.
Sign #1: The "Perfect" PR Description
If the Pull Request description is three paragraphs long, perfectly formatted, and uses words like "furthermore," "additionally," and "in conclusion"—it was written by AI.
The AI Description
"This PR implements the user authentication flow. Furthermore, it addresses the edge case where the token is invalid. Additionally, unit tests have been provided to ensure robustness."
The Fix: Ask them to summarize the change in one sentence during standup. If they can't, they didn't write it (or didn't read it).
Sign #2: Variable Name Amnesia
You ask, "Why did you name this variable dataList?" and they stare blankly.
When you write code yourself, you agonize over names. When you accept an AI suggestion, you gloss over them. If your dev doesn't know why a variable is named what it is, they didn't pick it.
Sign #3: The "Try-Catch" Blanket
AI loves to wrap everything in a try-catch block. It's a safe, defensive pattern. But in a production app, you often want things to crash (so you can fix the root cause) or bubble up to a global handler.
try {
await saveUser(user);
} catch (error) {
console.error("Error saving user:", error); // AI's favorite line
return null;
}The Fix: Ask, "What happens if this fails silently?" Force them to think about the error state, not just suppress it.
Sign #4: Over-Commented "What"
We covered this in our previous article, but it's a huge tell. If you see comments like // Define constant or // Return result, it's a sign they are pasting code without filtering the noise.
Sign #5: The "It Works" Defense
When you point out a weird pattern, and their only defense is "But it works," that's a red flag. It means they value the output but don't understand the process.
How to Mentor in the AI Era
Don't be the "No AI" manager. Be the "Explain It To Me" manager.
- Code Reviews: Ask "Why?" more than "Fix this."
- Pair Programming: Watch them code. Do they reach for the tab key immediately? Challenge them to turn off Copilot for an hour.
- Vibe Checks: Use tools like Vibe Code Detector to show them the difference between "working code" and "human code."
The best junior devs today aren't the ones who don't use AI. They're the ones who use AI to learn, not just to ship.