Disclaimer: This was a mock interview conducted with a friend who is a Senior Engineer at a FAANG company. I did not actually waste a recruiter's time.

> THE_SETUP

[INTERVIEWER (Zoom)] -> [Audio Loopback] -> [Whisper AI] -> [Text] -> [ChatGPT-4] -> [My Screen] -> [ME READING IT]

The goal: Answer every question using ONLY what ChatGPT generates. No thinking allowed. I am just a text-to-speech engine for the AI.

> ROUND 1: CODING CHALLENGE

INTERVIEWER: "Okay, let's start with a classic. Given a binary tree, I want you to invert it. But do it iteratively, not recursively." INTERNAL: Typing furiously... prompt sent. ME (Reading ChatGPT): "Sure. To invert a binary tree iteratively, we can use a queue to perform a Breadth-First Search. For each node we visit, we swap its left and right children." INTERVIEWER: "Sounds good. Code it." INTERNAL: ChatGPT generates code. I type it out as if I'm thinking.
def invertTree(root): if not root: return None queue = [root] while queue: current = queue.pop(0) current.left, current.right = current.right, current.left if current.left: queue.append(current.left) if current.right: queue.append(current.right) return root
INTERVIEWER: "Clean. What's the time complexity?" ME: "O(n), where n is the number of nodes, because we visit each node exactly once." INTERVIEWER: "Great. You nailed that."

> ROUND 2: SYSTEM DESIGN (THE TRAP)

INTERVIEWER: "Design a URL shortener like Bit.ly. It needs to handle 100 million writes per day." INTERNAL: Prompting... "Design Bitly, scale 100M writes." ME: "First, let's estimate the storage. 100 million writes * 365 days * 10 years... we need about... uh..." INTERNAL: ChatGPT is hallucinating math. It says 3650 TB. That seems high. I say it anyway. ME: "We need 3650 TB of storage." INTERVIEWER: "Are you sure? A URL is like 100 bytes. 100M * 100 is 10GB a day. Over 10 years that's maybe 36TB. You're off by factor of 100." ME: "Ah, yes. I... carried the zero wrong." INTERVIEWER: "Okay, how do you generate the unique keys?" ME (ChatGPT): "We can use a Base62 encoding. Or we can use a UUID generator. Or we can use Zookeeper to manage ranges." INTERVIEWER: "Which one would you choose and why?" INTERNAL: ERROR. ChatGPT gave me a list, not a decision. ME: "I would choose... the best one." INTERVIEWER: "..."

> ROUND 3: BEHAVIORAL (HALLUCINATION)

INTERVIEWER: "Tell me about a time you had a conflict with a coworker." INTERNAL: I don't have a story. I ask ChatGPT: 'Generate a fake story about conflict with a coworker.' ME: "Sure. Once, at my previous job at... [Insert Company]... I was working on the 'Quantum API'. My coworker, Bob, wanted to use Python. I wanted to use Rust. We fought. I initiated a duel." INTERVIEWER: "A duel?" ME: "Metaphorically. We A/B tested the performance. Rust won. Bob cried. We are friends now." INTERVIEWER: "Bob... cried? Over Python?" ME: "Yes. He is very emotional."
RESULT: REJECTED

> ANALYSIS_LOG

Where it Excelled: Coding. LeetCode questions are its training data. It solves them faster than any human.

Where it Failed: Nuance. In System Design, it throws out every buzzword (Kafka, Redis, Sharding) without understanding why. In Behavioral, it sounds like a sociopath trying to emulate human emotion.

The "Lag" Problem: The 3-second delay while I waited for the text generation was noticeable. I had to say "Hmm, interesting question..." about 50 times. The interviewer knew something was up.

Conclusion: You can use AI to cheat on the coding test. You cannot use it to cheat on the job.