What AI can't replace in a job interview

There’s a hot take making the rounds: “If AI can solve your interview question, the interview was pointless anyway. It’s not like you can’t use AI on the job.”

I think it gets the logic exactly backwards.

That take assumes the interview was testing whether you could solve the problem. It never was. I sit on the offer side of the table, and I can tell you: once AI flattens the solving layer, what’s left exposed is precisely what we were always trying to measure.

So should you use AI in your interview? Yes, use it. But I’m watching how you use it: to think with, or to think for you. That single line decides whether I score you up.

What it’s actually testing

A problem with a clean answer in the AI search bar isn’t a filter. It’s a prompt.

What I’m actually listening for is what’s behind the answer: the three things AI can’t hand you.

The first thing that shows up is how you frame the problem. Two candidates get the same prompt. One grabs a framework off the shelf. The other pauses, turns it around, asks “what are you actually trying to optimize here, in this scenario?” and works to align with me before writing anything. The second one hasn’t written a line yet, and I’ve already started giving them credit.

Next: when there’s no clean answer, how do you make the trade-off? Most real problems at work leave you with A and B both imperfect. Which one do you bet on? Don’t look at AI for this one. It doesn’t have your context; every answer it gives you sits on assumptions it quietly made on your behalf. The stance has to be yours.

The last one: what you do after the call. If the bet lands, how do you scale the result? If it doesn’t, how fast do you notice, and how fast do you turn? AI can’t carry this step either. It will generate a hundred options for you, but the decision to change course is yours alone.

Three things AI-era interviews actually test

1. How you frame the problem
2. Where you stand in the ambiguous middle
3. What you do after the call: scale it, or turn fast

So all I watch is one thing

Amplify, or substitute.

Amplify looks like this: rapid trial and error, research, compressing the grunt work, and spending every ounce of freed-up thinking on those three things above. I rate that up, because that’s simply what work looks like from here on.

Substitute looks like this: the question goes straight to AI, no framing of your own, no trade-off logic, an answer that points the right way but can’t survive a single “why.” That’s a no-hire. I’m not hiring a relay for an AI. If I wanted a relay, I’d hand the work to my own agent and skip the headcount.

Same visible action, “used AI in the interview.” Two completely different outcomes.

Why this is the only thing I score you on

AI has compressed the time it takes to produce a credible-looking answer down to seconds. That used to be the work. It’s not the work anymore.

What’s left is the part that doesn’t speed up:

→ Can you tell which half of the problem is the real one? → When the data is muddy and the stakeholders are split, do you have a position? → After you make the call, can you scale what worked and turn fast on what didn’t?

These don’t test knowledge. They test judgment. And judgment is what an interview was always trying to surface, before AI made everything else look like a solved problem.

I build agent workflows myself, and I watch them absorb every task that has a findable answer. Sitting on the other side of the interview table is what made this click for me: AI took the half of my work you could search. The other half, the half with no answer key, where you make the call and live with it, is where offers actually come from.

I keep meeting candidates who are brilliant at producing answers and not as good at deciding.

That’s the gap I’m hiring against.

---

If you’re interviewing right now, or you’re the one sitting across the table: I’d love to know what line you draw.