Let's Use AI to Generate Code! - A Guide to Spending 5 Weeks Debugging

At first glance, AI-generated code looks fantastic. It’s well-formatted, follows best practices (or looks like it does), and will likely pass a code review from an overworked senior dev. But underneath the synthetically perfect exterior lies a monster—one that hides, patiently waiting until it is merged into production.

But this isn’t a new problem. We’ve been doing this for years. People trust AI-generated code because it’s delivered with such confidence. Kind of like a GPS that insists the river in front of you doesn't exist, AI-generated code is rarely as solid as it appears. And when it fails (and it will fail), debugging it is hell. It’s like trying to untangle Christmas lights, in the dark, with stray cats “helping” you.

The Risks of AI Hallucinations

One of the biggest dangers of AI-generated code is hallucination—where it confidently invents code, APIs, or libraries that don’t actually exist. And let’s be honest, that’s a kind way of saying it lies and makes stuff up but delivers it with such confidence. Sometimes that means it will just pretend a function exists—at least the compiler catches this one. But other times, it can write security protocols that might as well be ROT13 but worse.

Fake API Endpoints & Nonexistent Libraries

AI loves to suggest calls to libraries that straight-up don’t exist but sound like they should.

“Just import futuristicML.OptimizedAuth() - it’s totally real, bro, trust me!”

Dangerous Security Practices

AI-generated authentication? You mean the backdoor generator!

One model confidently wrote:
passwordHash = hash(password, "MD5");

Another suggested storing encryption keys inside the code… in plaintext! (Though in fairness, I’ve seen human devs do this too.)

Endless Loops & Broken Logic

100% CPU usage and an unresponsive server? Turns out that genius AI model forgot that recursive functions need a base case.

Silent Failures That Gaslight You

catch (Exception ex) { return null; }

Why is this returning null? No errors in the logs.

Yay, it’s debug-day!

This isn’t just the theory of a developer terrified of losing his job to AI. People are already suffering in production because of AI-generated code. Just look at Stack Overflow or Reddit, and you’ll see long discussions trying to debug various issues—later learning a core element was written by AI and they don’t really understand how it works. And, of course, the now infamous case of lawyers citing fake cases invented by AI (source: Reuters - incidentally I don't often see AI quote it's sources!).

So Should We Ever Use AI-Generated Code?

Well… sometimes. AI can be a powerful tool—just remember it’s a tool, and therefore needs to be used correctly. It’s a helper, not a replacement for thinking.

It makes sense to use AI when:

• Generating boilerplate code. Let’s face it, you’ll just be copying and pasting otherwise.
• Writing small, self-contained utility functions that you can easily verify. You are writing unit tests, right?
• Using autocomplete and code suggestions (but remember to check them!).

But you really shouldn’t use it for:

• Writing business-critical logic (unless you like surprises).
• Anything security-related (unless you like lawsuits).
• Code you don’t fully understand (unless you like debug-days).

We’ve Been Here Before

AI-generated code didn’t just show up with ChatGPT. We've seen a long history of "No Coding Required" tools promising simplicity but delivering terrifying spaghetti code.

Are you old enough to remember Dreamweaver or HoTMetaL? They promised to generate HTML for you, but the result was a bloated, unmaintainable mess, filled with unnecessary <font> tags. If you don’t remember this, you’re either a sweet summer child, or you’ve repressed it well.

History repeats itself—this time, AI is the new overhyped tool. And just like Dreamweaver-era HTML code, AI-generated code will haunt future developers for years.

AI Is a Tool, Not a Developer

At the end of the day, AI is just a tool—like a debugger, a linter, or Stack Overflow. And just like Stack Overflow, blindly copying and pasting code from AI is a bad idea.

I mean, you wouldn’t just copy and paste code from Stack Overflow without fully understanding it first, right? …Right?

AI isn’t here to replace developers—it’s just going to flood the industry with bad ones. And let’s be honest, we already have enough of those! But a good developer knows how to use their tools effectively.

In an ideal world, thorough code reviews, automated unit tests, and functional testing should catch AI’s mistakes before they get merged—but very few of us live in that world.

So if you’re going to use such a powerful tool, make sure you use it correctly, for the right jobs, and be safe. Bonus points if you make a note when using AI-generated code so reviewers can give it some extra scrutiny.