Halfway through a technical interview, I asked a candidate a simple follow-up question. His reply helped me realise I wasn’t talking to him. I was talking to an AI, with him as a middleman.
AI has changed technical hiring: A take-home challenge is easier to fake, and a polished answer no longer proves someone understands anything. But banning AI isn’t the answer either, because the people who use it well are exactly who you want to hire.
In this post I’ll share my insights regarding how I handle AI in the hiring process, how I spotted a fake candidate in real time, and what you can do to spot one too.
1. Technical Hiring
Take-home challenges have long been the filter for who moves on to the next stage. With AI, anyone with little knowledge can now solve them, which quietly lowers the bar.
Why not ban it
As with any other tool, I don’t believe blocking its usage is the right answer, especially when its expected to be leveraged for that role, plus it can help produce high-value results. But I also agree with a statement often shared:
“Fully delegating to AI makes us dumber, because we stop exercising our brain muscle.”
Both extremes are undesirable, so the solution is probably somewhere in the middle.
Using AI well takes expertise
Left unguided, AI can solve a problem in ways you never wanted, even causing collateral damage.
In the context of take-home challenge, I delivered one hacking challenge as a container where it was strictly forbidden to get in via docker exec - You’d be surprised how many reports I received that did exactly that.
Fully delegated work
There is so much AI-generated content out there that fully delegated work is often easy to spot, and in my opinion, that’s undesirable. If you’re delivering a written report, show a bit of yourself and use your creativity. It’s what will make you stand out from the flock.
Last but not least, always verify your deliverable for consistency of ideas and alignment with what has been asked for the task at hand.
AI has removed the cost of looking competent. It did not remove the cost of being compentent
Technical Interviews featuring AI
Even though AI should be allowed for solving technical challenges, during the interview I usually ask:
- Did you use AI to solve the challenge?
- Walk me through when and how you actually used it.
- Was there a specific step you found harder than others? Why was it harder?
This takes the taboo out of AI usage, and it shows me how the candidate leveraged it. Most importantly, it shows whether they used it to understand the core principles of a problem they found hard, and learned how to solve it.
2. Spotting Fake Candidates
A few months ago, during a technical interview process, there was one candidate that brought AI usage to a whole new level.
The following are the multiple stages of how I went from “something feels off” to “this interview is over”.
1. The eyes
I usually start by putting candidates at ease with casual questions. While we chat, I watch how they respond, including their expressions on camera.
Something felt off straight away. The candidate’s eyes were never on the camera. They moved sideways, left to right, the way someone do when reading. The casual questions didn’t land smoothly either. He seemed to expect only role-related questions, and asking questions to be a bit creative or authentic made him visibly uneasy.
None of this proves anything on its own. People read notes, use a second monitor, or are just nervous but I took note of it anyways.
2. The question that broke the script
I started with the three homework questions from earlier. Then, because I had read his take-home challenge report, I picked a mid-step of the first challenge and asked:
“There is this step of the first challenge where you did X. Why did you do it that way?”
His reply:
“Should I refer to the homework report?”
I was confused. We had been talking about that report for minutes. Why would he ask that? I followed up:
“What do you mean?”
I watched him realise it was a nonsensical question to ask. He ignored it and carried on answering.
And he didn’t answer my question. He started from the beginning of the challenge and narrated the whole thing, non-stop. While he talked, I opened the report and noticed that everything he said was a slightly different flavour of what was already written. Zero entropy, no pauses, no creativity and no human communication.
At this point I was nervous. Having a live interview that’s just someone narrating AI-generated responses was crazy enough, but confronting it was a first for me.
I was 80% sure there was an AI setup intercepting my questions and he was reading its replies. But I was so incredulous that I decided to keep going…
3. Answers that were too complete
Now my brain was searching for every AI usage signal. In the technical questions, his answers were all very “complete”, extending well beyond the scope of what I asked, with examples I’d never asked for and did not align well.
Here are two examples from my notes:
| I Asked | My Notes |
|---|---|
| Can you explain Hashing vs Obfuscation | He says Hashing is not good use because it provides random characters and obfuscation is the way to go because thats what you use in HTTP settings and Java - this does not make any sense, and when I asked follow up he just stared at the screen and got nervous. |
| Can you explain Encryption vs Encoding | “This guy looks like he is reading from somewhere.” |
To test it further, I used generic follow-ups like:
- “Please elaborate on that.”
- “How does that relate to the topic at hand?”
- “Yes, and then?”
AI’s default behaviour is to keep expanding on whatever the context is. It rarely admits it doesn’t know, and it rarely keeps things short. Every answer he gave grew longer, with non-stop narration and technical examples that had nothing to do with the domain of my question.
Eventually I could not make any sense of what the candidate was saying and how it related the initial question, so I said:
“I think it’s better we stop this interview.”
He agreed. I closed the call.
Closing Thoughts
That interview changed how I think about hiring. Assessing skill and culture fit is no longer enough: I also have to confirm that the person in front of me is the one answering, and that they use AI to grow, not to hide behind it.
To close, here’s a Tim Ferriss reel I came across not long ago:
“AI is an accelerant, it makes more of whatever is. If you do not first identify the critical few, AI simply multiplies the useless, it multiplies the unimportant, the trivial, the distracting.”
I’ve turned what I learned into a small cheat sheet you can keep for future reference. Get it for free here: [link].