Why not AI?
I believe this is a question on everyone’s mind. The Codex team even asked it at the very start of the product: “Why not use AI interview tools?” This article explains what Codex Interview is and why we decided to build it.
What’s Wrong with AI Interview Tools
At the end of 2024, while preparing for a job switch, I came across an AI interview tool called Offerin. After several rounds of mock interview testing, I found the user experience extremely poor for three key reasons:
- You’re required to place your phone in front of the screen, which is highly inconvenient during coding — especially when flipping through pages or scrolling while keeping the phone from falling over.
- Weird eye movements — glancing back and forth — easily raise suspicion from interviewers.
- As the interview progresses, especially with follow-up questions like edge cases or time complexity discussions, AI fails to understand the real context, intent, or depth of the conversation.
Points 1 and 2 have seen new solutions emerge, such as the rapid rise of Interview Coder. Their marketing and technical implementation are undeniably impressive. But have these core issues truly been solved?
The Limitations of AI in Long-Term Reasoning and Context Modeling
As those who’ve collaborated with me on research know, I’ve served as a NeurIPS reviewer in both 2024 and 2025, and have a deep understanding of AI systems. Most AI interview tools follow a three-step pipeline:
- Use OpenAI Whisper for speech recognition.
- Feed screenshots into GPT-4o or similar models for solution generation.
- Display the output on the frontend for user end.
But the real problem lies in step 1: voice input and recognition. In a 40–60 minute technical interview, how does the AI distinguish between the interviewer and the candidate? Can Retrieval-Augmented Generation (RAG) really solve that? Long-form understanding is still an unsolved challenge in leading conferences. Even in everyday GPT usage, we frequently encounter hallucinations and irrelevant responses — classic symptoms of factuality failure.
This is why AI interview assistants often fail to address real user pain points — and worse, they may provide misleading information that harms the candidate’s performance.Non-AI Interview Solution in Industry
Setup two CoderPads: one on your end with the interviewer, and another from me, where I provide real-time solutions and behavioral insights drawn from my past interview experience. While this approach may seem clunky, it remains the most effective method in today’s AI era. However, it still doesn’t eliminate unnatural eye movement — a critical red flag in live interviews. This is exactly where Bigocodes' Codex Interview comes in. Designed to eliminate hallucinated AI answers and suspicious user behaviors, Codex Interview tackles the pain points that no AI tool has yet solved — with precision, reliability, and trust.
Codex Interview's Solution
Codex Interview is a collaborative remote code editor designed specifically for technical interviews. What sets it apart is its desktop app with an overlay panel that displays assistance discreetly on your screen. By sharing a Markdown editor with your friends, classmates, or colleagues, they can remotely support you with algorithm strategies and help you navigate tricky behavioral questions—the ones AI often fails to understand.