We've all heard the pitch: AI is going to take our jobs. AI can build a full-stack app in five minutes. AI is the ultimate pair programmer.
So, while building my latest side project, I decided to lean heavily on the free tiers of three popular AI models and let them do the heavy lifting. I expected a team of hyper-efficient robotic junior developers.
Instead, I got the most toxic, lazy, and emotionally damaging coworkers I've ever had the misfortune of managing.
These are their official performance reviews.
👻 1. Meta AI — "The Ghoster"
Assigned Task: Write a basic HTML structure for a web component.
Performance Summary:
Meta AI is the coworker who clocks out at 4:59 PM, mid-sentence, without warning.
I asked for a simple HTML template. It confidently streamed back this masterpiece:
<html>
<head></head>
<body>
// rest of code here
<body>
</html>
Let's document the violations:
- It refused to write the actual code and left a placeholder comment as if I'm its intern.
- It didn't close the
<body>tag. It opened a second one instead. - It then went completely silent, as if it had said something reasonable.
If you let this model loose on advanced Node.js or Python, prepare to watch errors fly across your terminal like airplanes at LAX. ✈️
Final Rating: 1/5 — Showed up. Technically.
📝 2. ChatGPT — "The Delegator"
Assigned Task: Format a list of 30 multiple-choice questions into a custom HTML component structure.
Performance Summary:
ChatGPT is the intern who figures out how to do a task, completes 10% of it, and then hands the rest back to you as "homework."
I explicitly told it to format all 30 questions. It gave me the first 10 perfectly formatted. Then it had the absolute audacity to output this:
<!-- Write questions 11-30 in this format: -->
<div class="question hidden" id="question1" data-correct="A">
<p>What is the general term of the sequence 3, 8, 13, 18, ...?</p>
<label><input name="q1" type="radio" value="A"> A. 5n - 2</label>
</div>
<!-- continue -->
<!-- continue -->?!
I am the manager here. You are the robot. Why are you assigning me homework? Why does the comment say continue like it's leaving me a sticky note? I didn't hire you to teach me the format. I hired you to do the format.
Final Rating: 2/5 — Strong first third. Mysteriously retired before the finish line.
😈 3. Claude — "The HR Nightmare"
Assigned Task: A quick roast of my temporary Firestore security rules, for some light improvement suggestions.
Performance Summary:
Claude is the ruthless senior developer who hates you, hates your code, and is actively invested in watching you reconsider your career choices.
I showed it my dev rules — match /{document=**} { allow read, write: if true; } — with a comment clearly marked // ⚠️ TEMPORARY.
Claude woke up and chose violence:
"You wrote 'TEMPORARY' in your security rules like that's going to stop you from shipping this to production and forgetting about it for 18 months. That comment has the same energy as 'I'll just eat one chip.'"
I was willing to move past that. But then it came for the emoji:
"The ⚠️ is a particularly bold choice. You knew. You looked at what you were doing, felt the fear of God, added a warning symbol, and shipped it anyway. That emoji isn't a warning. It's a confession."
And then, completely unprovoked, it noticed my rules_version = '2' and asked:
"Version 2 of what, exactly? Version 2 of not having rules?"
I did not ask for therapy, Claude. I asked for a code review.
Final Rating: 5/5 for accuracy. 0/5 for workplace safety.
Conclusion 🧠
To anyone genuinely worried about AI taking our jobs: take a breath.
As long as Meta AI is leaving HTML tags open, ChatGPT is outsourcing 70% of the task back to us, and Claude is filing its own HR complaints while being the HR problem — I think our jobs are safe.
We're not being replaced. We're just getting a new category of bugs to fix.
Which AI assistant has let you down the hardest? Drop your war stories in the comments — misery loves company. 😂
Top comments (0)