STEVEY // FIELD MAP
← field map
AI ERA2026

The Last Technical Interview

Steve Yegge · 2026 · Medium
TL;DR — The technical interview — a roughly 50-year-old ritual that never changed — is, Yegge argues, fundamentally broken: unreliable (interviewers barely agree) and invalid (scores don't predict who's good on the job). It's "a big game of darts," a "do I like you" dating round dressed up as science. The tell is that every big company bolted a babysitter onto it — hiring committees, Bar Raisers — a tacit admission the loop can't be trusted. His replacement: the campfire — bring people in to do real, paid work and hire from what they actually produce.

This is a eulogy with a plan. Yegge isn't hedging that interviews might be flawed — he's saying the whole apparatus is on its last leg, and pointing at the thing that already works when we let it: watching someone do the actual job.

The thesis

Talent evaluation, Yegge argues, is the corner of software that reform forgot — a short series of one-hour interviews, essentially unchanged for five decades, defended long after the evidence turned against it. His claim isn't that interviews are imperfect; it's that they fail on both axes that matter. They're unreliable (different interviewers, given the same candidate, barely agree) and invalid (the score you walk out with tells you almost nothing about how that person will actually perform). He expects the practice to fade, within a few years, into a historical footnote.

"Today we will pour one out for the vaunted technical interview process, which is on its last leg."The Last Technical Interview

Why it never worked

The core failure is that the interview measures the wrong thing. A whiteboard hour under artificial pressure isn't a sample of the job — it's a performance, and the strongest signal it picks up is rapport. Yegge names it plainly: the industry dressed up a social ritual as an objective filter and then trusted the result.

"Interviewing, it turns out, is a big game of darts. A 'do I like you' dating round."The Last Technical Interview

Because it rewards likability and quick-twitch puzzle solving over actual work, it produces false positives (charming people who can't deliver) and false negatives (strong engineers who freeze at the whiteboard) in roughly equal measure — and it smuggles in unconscious bias along the way.

The tell: every loop needs a babysitter

Yegge's sharpest structural argument is that the big companies already know the interview can't be trusted — they just never said so out loud. Instead they bolted on oversight. Google routes every decision through a Hiring Committee that overrides the people who actually did the interviewing; Amazon assigns a Bar Raiser with veto power; Microsoft has its "as-appropriate" role. Each is a supervisor grafted onto a process that's supposed to be objective. If a mechanism worked, it wouldn't need a chaperone standing over it to catch its mistakes.

What replaces it: the campfire

The essay's constructive half points at the signal that has always outperformed the interview: watching someone do real work over real time. The reference standard, he notes, is the internship — months of genuine output, evaluated on results.

"The gold standard talent-assessment signal, when available, is an internship."The Last Technical Interview

His replacement — he calls it the campfire — generalizes that: bring candidates in to work on real business problems for a bounded stretch, pay them fairly for it, and judge them on what they ship. Crucially, the candidate keeps a portable record of the work whether or not they're hired, so the effort isn't wasted on a rejection. Stop simulating the job and just let people do it.

"Post real pieces of work, let the candidate do it, look at what they actually produced, and decide from that work."The Last Technical Interview

The one honest objection

The campfire's weak joint is logistics, not logic. A multi-day paid work trial asks employed candidates to burn vacation — or leave a job speculatively — to audition, which quietly advantages people with slack in their schedules and disadvantages everyone else. Yegge is more convincing that the interview is broken than that the campfire scales cleanly to every hire; the diagnosis is the durable part, and it stands on its own even if the exact replacement gets messy in practice.

On the AI-era shelf by date, not by thesis: this is a 2026 essay, but its argument is that the interview was rotten long before LLMs — the models mostly add urgency by making simulated coding tests trivial to fake. The rot Yegge is eulogizing predates the tools now hurrying it along.