Some people relax on weekends. I, apparently, like to build small tools that roast my code.
This past weekend, I decided to make a fun (and mildly painful) experiment: an AI-powered code reviewer that takes my code, analyzes it, and gives feedback — no sugarcoating.
It was part tech challenge, part personal dare.
🤖 The Setup
I used the OpenAI API and a simple Python script. The input: raw code from my Laravel or Python projects. The output: brutally honest feedback from an LLM with no chill.
My system prompt?
“You are a senior developer who reviews code honestly and directly. Be concise and point out bad practices, if any.”
I didn’t realize how much I’d regret that line.
🔍 What It Caught
Here are a few things the AI pointed out (yes, real examples):
-
“Unclear variable naming. Avoid
xanddatain larger scopes.”
Ouch, fair. -
“You can refactor this into a reusable function.”
True. I was lazy. -
“No input validation. This is risky.”
Fine. You’re right. Stop yelling.
Some of it was repetitive. But a lot of it was actually helpful — especially when I pasted messy test snippets or older functions I’d half-forgotten about.
🧠 What I Learned
-
AI can spot low-hanging issues fast. Things like missing comments, repetition, or inconsistent naming? It’s all over them.
-
Context is everything. The model doesn’t really understand your codebase. It guesses. So, sometimes, it suggests changes that don’t apply.
-
Tone matters. The more aggressive I made the prompt, the funnier (and harsher) the results got.
💡 Could This Be Useful?
Honestly? Yes. It won’t replace a human reviewer — not even close. But it can be a first pass before asking your senior dev or mentor. And for solo builders like me, that’s valuable.
Also: it’s hilarious.
🧨 Try It Yourself
If you want to try something like this:
-
Get your OpenAI API key
-
Write a simple prompt in Python
-
Paste your code and see what happens
Just… maybe don’t ask it to be brutally honest unless you’re ready to be humbled.