Manual Override 🚗

My mother taught me to drive. Fully automatic, from day one. I have never in my life operated a clutch.

My father-in-law is a gearhead in the truest sense — the kind of guy who mods motorcycles in his garage for fun. A few years back he wheels one out, a bike he'd built with a sidecar, and asks if I want to take it for a spin. I tell him I've never ridden a motorcycle in my life. He waves it off. Throws me on. Tells me to slowly turn the right handle.

The engine screams. He's yelling something about engaging the gear. I have no idea what that means. He says something about the clutch. Still nothing. Finally, in what I can only assume was pure desperation, he shouts squeeze the silver thing.

I squeeze the silver thing. The bike goes flying in a circle — sidecar and all, me hanging on for dear life — while he stands in the yard watching his motorcycle do donuts with his son-in-law strapped to it.

He wasn't frustrated because I couldn't ride a motorcycle. He was frustrated because he'd assumed I already had something I simply never built. In his head, "he knows how to drive" and "he knows how to work a clutch" were the same fact. They weren't. I'd been taught on automatic. The reflex he expected me to have on reflex — engine, gear, clutch, throttle, in that exact order, under pressure — had never been installed in the first place. No amount of shouting instructions mid-motion was going to summon a muscle memory that was never there to begin with.

I think about that morning every time I read one of these interview-standards memos now.

I work at a car marketplace. I did not pick this metaphor for the pun. The pun picked me.

The Memo 📋

An internal group responsible for interviewing standards sent out guidance last week on handling candidates suspected of using an LLM during a live coding round. The instructions were specific: interrupt gently, ask the candidate to walk through their thinking before typing further, watch for "unnaturally generic spoken explanations," flag it to the recruiter and hiring manager afterward regardless of how the interview ends.

The stated reasoning was fairness. Candidates who don't cheat shouldn't lose to candidates who do. Fine. Reasonable, even.

Here's the problem. The same organization that wrote that memo has, over the past year, told its engineers — in performance conversations, in roadmap reviews, in the actual language used to describe what "senior" now means — to stop writing code by hand wherever an agent can do it faster. Not "use it when it helps." Stop coding. Rethink your role. The muscle you're being graded on at the door is the muscle you're being told to retire once you're inside.

Fine, Prove You Can Drive Stick 🧠

I want to steelman this before I take it apart, because the lazy version of this post is "haha hypocrisy" and that's not actually interesting.

The charitable read: an interview measures a baseline. You want to know a candidate can reason about a problem, hold a data structure in their head, explain a trade-off without outsourcing the thinking — before you hand them a tool that can fake all of that convincingly. Test the fundamentals first. Automate later. That's not insane. Driving schools still teach you to parallel park before you buy a car with a camera that does it for you.

What it looks like

• Live coding rounds with AI assistants explicitly disallowed
• Interviewers trained to pivot with process questions when they smell AI in the room
• Feedback flagged in Greenhouse, escalated to the recruiter, no ambiguity in the paper trail

The upside

It does, genuinely, protect signal. An interviewer watching someone read code off-screen line by line isn't measuring anything real. The instinct to want your own voice, your own reasoning in the room isn't the problem.

Where it collapses

The baseline defense only holds if the company still believes the baseline matters after the interview. Mine doesn't. If hand-written code is legacy behavior on the job, then testing for it at the door isn't measuring a foundation — it's measuring a skill the company has already filed under deprecated. You're not checking if someone can drive before you sell them an automatic. You're checking if they can drive stick, for a job where the pedal's been removed.

Nobody's Driving 🌆

Here's what actually gets me, and it's not the contradiction. Contradictions I can respect — at least they imply someone made a call. What I'm looking at is dumber than that.

Two groups in the same company issued two internally consistent, mutually incoherent policies, and neither one has apparently talked to the other. One team is optimizing for "we sound AI-forward to the market and the roadmap." The other is optimizing for "our hiring signal doesn't get contaminated." Both are defensible in isolation. Neither one checked what the other was doing.

If this were a strategy — if leadership had actually decided interviews measure who you were, the job measures who you're becoming — I'd disagree with it, but I'd respect that someone thought it through. It's not that. It's two memos from two silos that happen to describe the same engineer in opposite tenses.

In other words: this isn't a company with a coherent, if cynical, philosophy about engineers. It's a company that hasn't decided what an engineer is anymore, and sent both memos anyway.

The Assumption Gap 💭

But the sidecar story isn't really about hypocrisy. It's about what happens to the person caught in the middle of it — and there are two different people caught in it, not one.

There's the engineer who learned to drive stick years ago and has been handed an automatic for the last two years. That reflex doesn't stay frozen in place from disuse. It rusts. The clutch feel you had is not the clutch feel you have now, because nobody keeps a reflex sharp by not using it.

And there's the newer engineer — the one who came up entirely AI-first, who learned to build software the way I learned to drive: automatic, from day one. Nobody handed them a manual car to practice on, because for the entire span of their career, nobody around them was driving one either. They were never taught the clutch. Not "let it atrophy" — never built it.

The interview panel doesn't distinguish between these two people. It treats "can this person write code" as one fact, the same way my father-in-law treated "can this person drive" as one fact — assuming that competence in one mode implies latent, dormant competence in the other, ready to surface the moment you shout the right instruction at them mid-motion. Just engage the gear. Just explain your reasoning without the tool. As if the reflex was always in there, idling, waiting for permission.

For some candidates that's true — it's atrophy, and it comes back with practice. For a growing number, it isn't true at all. It was never installed. And no amount of gentle interruption, no pivot question asked with good intentions in the middle of a live interview, is going to summon a muscle memory that was never given the chance to form.

That's the part that should scare hiring teams more than cheating does. You can catch someone using a tool. You can't shout someone into a reflex they were never taught.

The Interview I Actually Believe In

I wrote a whole post about what a good technical interview should feel like — be a good host, make it a conversation, ditch the courtroom drama, remember you're not trying to trap anyone. I believe every word of that. Still do.

This memo is the opposite instruction, issued by the same building. Interrupt them. Watch for tells. Flag it to the recruiter regardless of outcome. That's not hosting a conversation. That's standing in the yard, watching someone spin in circles, and being certain they're the one who did something wrong.

You can't keep taking the clutch away — or never installing it in the first place — and then grade someone on the biting point.

Pick one. Sell stick. Sell automatic. But stop building a dealership where the showroom sells one car and the test drive demands the other.