Last week I wrote about getting rejected from a job for "relying on AI" in the coding round — by an interviewer I'd watched read an AI answer off his own screen forty minutes earlier. I thought the story was about hypocrisy.
Then a reader replied with something sharper, and it reframed the whole thing for me.
He'd been through the opposite version. Told "please, no AI," so he stuck to it — and blanked on a manual code review of a file that couldn't even compile, because of external calls he was never shown. He walked out certain he'd be lowballed. And buried in his comment was the sentence that's been rattling around my head since:
The interview isn't for judging competence. It's for judging value — and for seeing how far they can degrade your perceived value before they name a number.
I read that and something clicked into place that my own rejection had only hinted at.
So I did the thing the rejection process never lets you do: I put it on the table and interrogated it. Not the interviewer — the machine. Two rounds, mine and his, treated as evidence about what the whole apparatus is really built to do. Here's what their hiring process is actually optimizing for.
The test I thought I was failing
Both of us walked into our rounds thinking the same thing: this is a competence test. Prove you can do the work.
That's the wrong frame. It's not a competence test. It's a pricing test.
The company already believes you can probably do the job — your résumé got you in the room, your GitHub is right there, they can read. What they don't know yet is the one number that decides whether hiring you is a good deal: how little you'll accept. And the interview is the instrument they use to find it.
Once you see it that way, a lot of otherwise-baffling interview behavior snaps into focus.
Why they need you to fail at exactly one thing
Here's the part that sounds cynical until you've lived it.
The ideal interview outcome for the company is not "flawless candidate." A flawless candidate is expensive. If you sail through every stage, you walk into the offer conversation with leverage, and you'll refuse a lowball because you have every reason to.
The ideal outcome is a candidate who is clearly good — good enough to hire — but who fumbled exactly one thing. Not zero (then they can't discount you). Not everything (then there's no offer). One clean, legible stumble. Because that one stumble is the whole negotiation:
"You're strong. Really. But you're not quite as senior as you think — remember that code review? So here's what we think you're worth."
And you, still stinging from the one thing you got wrong, nod and accept a number well under where you walked in. The stumble wasn't a bug in the interview. It was the product.
The tell is in the questions they ask
You want to know whether an interview is measuring your competence or pricing your floor? Listen to how they ask about money.
- "What are your salary expectations?" asks for your ceiling — the most you think you're worth.
- "How much do you currently earn?" asks for your floor — the least they can get away with.
Guess which one gets asked. Nearly every time, it's the second. They are not trying to find out what you're worth. They're trying to find out what you'll settle for — and anchoring the entire offer to a number from your past instead of the value in front of them.
That single swap tells you which game is being played. It was never "can you do this." It was always "how little can we pay you to."
Value in the room, priced off the floor
The reader's story is the whole argument in one interview.
Walk in and bring real value — spot the bottlenecks their own team hasn't named out loud, demo something that plugs straight into their stack and beats what they're running now, watch the interviewer get visibly taken aback by how much you see. That's the entire job. That's the thing that's supposed to be rare and worth paying for.
Then fumble one manual code review — reading a file cold that can't compile because of calls you were never shown — and that gets to veto all of it. Not because it revealed you can't do the work. Because it handed them the discount they needed.
Value shown at the ceiling. Priced at the floor. The stumble laundered a lowball into "honest feedback."
And here's the tell inside the tell: the code you were failed for reading? It was running deprecated patterns on their live service. You got marked down for not fluently reviewing code that their own production shouldn't be running. The abstinence test and the code-review test are the same test — both grade you on a skill the job itself no longer rewards, precisely because that's the skill you're most likely to visibly miss.
This is the same failure as the AI thing
When I put the two stories side by side — mine and his — they're the same shape.
Mine: rejected for using the tool everyone in the room was already using. His: rejected for not fluently performing a skill the company's own production had already abandoned. Both interviews measured the cheap, legible thing instead of the expensive, real one. Both optimized for a stumble they could point at, not the value they could actually use.
Because the stumble is legible. "He used autocomplete." "He couldn't read the file cold." Those go in a scorecard. Judgment doesn't. The bug I flagged unprompted — persist-before-ack, the one that locks a paying customer out with no record on the worst possible day — never made it onto anyone's scorecard, because you can't screenshot it and you can't discount someone for having it. So it got ignored in favor of the thing they could.
So what do you actually do about it
You can't fix their instrument. But you can stop mistaking their price for your worth.
- Know your floor and refuse to hand it over. When they ask what you currently earn, answer with what you're looking for. You are allowed to move the anchor from your past to your value.
- Treat the one stumble as their tool, not your verdict. You did not fail the interview because you blanked on cold code review. You "failed" because the interview needed you to fail at something, and that was the cheapest thing lying around.
- Have enough runway to walk. The only real leverage in a pricing negotiation is the credible ability to say no. If you can let them make the junior offer and keep it — and let the runner-up have the seat — you've flipped the game. For once, you're the selective one.
- Sell the value they can't screenshot. Make the bottleneck you spotted, the bug you caught, the thing-that-breaks-at-2am you predicted — make those the loudest part of the conversation. If they still price you off a compile error, that's your answer about them, not about you.
Why this is exactly how I build
Here's why this isn't just an interview rant.
The failure in both of those rooms was a process that measured the legible thing instead of the valuable one — that couldn't tell the difference between "stumbled on something easy to point at" and "can't actually do the work." That is precisely the mistake I refuse to build into an agent.
I never let the thing that produces the work be the thing that grades it, and I never let the most legible signal stand in for the most valuable one. There's an author agent that writes the diff, a separate skeptic agent whose only job is to refute it rather than admire it, and a human on the merge button who can still see the blast radius — the value and the risk the model can't screenshot. The point isn't to find the one flaw and discount everything around it. The point is to judge the actual output. That separation between author, skeptic, and human is the whole shape of xenition, the agent platform we build.
An interview that hunts for your one stumble so it can price you off your floor is running the same broken evaluation I spend my days designing out of software. The company that prices you at your ceiling — that pays for the bottleneck you spotted instead of discounting you for the file you couldn't read cold — is rare. But they exist. And the only way you ever get to work for one is if you've kept enough self-respect to turn down the ones who don't.
Honest question for the comments: have you ever felt an interview go looking for your one stumble — the moment where you could feel them setting up the discount? Or been on the hiring side and watched a process reward the legible miss over the real value? I want to know how common this actually is. 👇
(If this landed, a ❤️ and a 🔖 help — and tell me the moment you realized they were pricing your floor, not testing your ceiling.)
Top comments (0)