The Last Technical Interview

Steve Yegge · article · 2026-05-29 source ↗ Reader 🖥️

AI Labor Disruption Hiring Signal Design Talent Density Trust

The traditional one-or-two-round technical interview has always been a statistically unreliable simulation of real work, and a combination of AI-driven strain (résumé floods, AI-assisted cheating, fraudulent applicants) and a growing supply of available senior talent is finally forcing the industry to replace it with paid, real-work trials — provisional employment models like internships, co-ops, and Yegge's proposed 'campfire' — whose results are made portable and reusable as a permanent candidate credential.

practical · mixed — Argumentative/practical: Yegge wants readers to abandon the traditional interview and adopt real, paid work trials. The case is built almost entirely through personal narrative and anecdote (Amazon Bar Raisers, the Google Hiring Committee calibration trick, Geoworks co-ops) rather than data or citation, with a short closing prescriptive sketch.

Structure

Part	Role
Personal authority framing (35 years interviewing, Amazon Bar Raiser, Google Hiring Committee) and thesis statement	context
'Remember That Time We All Fired Ourselves' — the Google HC calibration-exercise anecdote where the committee unknowingly voted to reject 2/3 of its own past interview packets	evidence
'Change is Taboo' — social/organizational account of why the industry hasn't fixed a process it knows is broken	context
'The Signal Problem' — a hierarchy of hiring-signal sources (résumé, phone screen, work samples, on-site interview) ranked by strength	conceptual framework
'Provisional Employment to the Rescue' — internships, the Geoworks 6-month co-op, University of Waterloo's 6-internship pipeline	evidence
Explanation of why provisional employment hasn't been adopted at scale (senior engineers wouldn't accept the risk) and why that equilibrium is now cracking (AI cheating, résumé floods, fraudulent candidates, a possible talent surplus)	evidence / turning point
'Let's Stop Simulating Work' through 'Where Do the Stamps Originate' — the 'campfire' prescription: paid multi-day real-work trials, work counted twice (value to employer + portable credential to candidate), interviewing reframed as a potential profit center	prescription

Key terms

Signal
The cumulative evidentiary basis an employer has for predicting a candidate's job performance; explicitly graded from weak (résumé) to strong (internship/co-op), with strength assumed to track roughly with duration and realism of the observed work.
Provisional employment Hiring Signal Design
A temporary work arrangement (internship, co-op, contract-to-hire) used specifically to defer a binding hire/no-hire decision until real on-the-job performance can be observed; explicitly distinguished from 'fractional' (part-time) employment.
Campfire
Yegge's proposed successor to the interview loop: bringing a candidate in for paid, multi-day work on a real codebase/ticket with a real team, as opposed to a one-day battery of simulated interviews. Coined in an earlier post of his ('the Anthropic hive mind').
Bar Raiser / As-Appropriate (AA)
A designated senior interviewer given veto power over an entire interview loop's hiring recommendation, instituted at Amazon and (per Yegge) Microsoft because individual interview teams cannot be trusted to reach good hiring decisions on their own.
Count Work Twice
The mechanism by which a candidate's trial work is used once by the employer (as both assessment signal and possibly usable output) and once by the candidate (as a permanent, portable, employer-signed record of what they did and how well they did it, regardless of hiring outcome).

Propositions

explicitThe standard interview process (a résumé pass, a phone/video screen, and four to six one-hour on-site interviews) cannot reliably determine whether a candidate can do the job.conf: high
explicitestablishedInterviewer judgments on the same candidate are highly inconsistent with each other and poorly correlated with the candidate's actual on-the-job performance.conf: high
supports ←Presentation: Getting Rid of LeetCode Interviews in the World of AI · p1 — Independent personal-experience evidence for interviewer judgments being poorly correlated with actual on-the-job ability — his own repeated interview failures despite deep, directly relevant expertise mirror the interviewer-disagreement/poor-correlation finding.
supports ←SWLW #711: The laws of this world (game theory), Should you still become a Software Engineer in 2026, and more. · p5 — Digest's retelling of the Google calibration test restates the source piece's claim that interviewer judgments are highly inconsistent and poorly correlated with on-the-job performance.
explicitcorroboratedTalent-assessment quality scales with the amount and realism of observed work: résumé < phone screen < work samples/credentials < on-site interviews < internship < multi-month co-op.conf: high
supports ←Presentation: Getting Rid of LeetCode Interviews in the World of AI · p4 — The Artsy hiring model (discuss actual prior work, take-homes, references) is a concrete instance of assessment quality scaling with the realism of observed work, landing between 'work samples/credentials' and 'on-site interviews' on that ladder.
explicitdevelopedProvisional employment (internships, co-ops, contract-to-hire) is currently the best available solution to the hiring-signal problem, because sustained real work generates dramatically more reliable signal than any interview loop.conf: high
extends ←The Case for Hiring Job Hoppers · p1 — The-last-technical-interview argues sustained real work (provisional employment) is the best available fix for unreliable hiring signal; this piece extends the hiring-signal-design problem by identifying a wholly different, already-available signal — candidates' prior mobility history — as more informative about likely on-the-job adjustment speed than conventional wisdom assumes.
explicitProvisional employment has historically been rejected at scale because competition for engineering talent meant senior candidates would not accept the risk and uncertainty of a multi-month trial role.conf: high
explicitcorroboratedThat equilibrium is now breaking because of converging pressures: AI-assisted cheating on assessments, résumé floods worsened by AI-generated writing, fraudulent applicants (e.g. North Korean IT workers posing as remote candidates), and — implicitly — a growing pool of engineers 'between gigs' with fewer alternative offers, making provisional roles newly acceptable to senior candidates.conf: high
supports ←SWLW #711: The laws of this world (game theory), Should you still become a Software Engineer in 2026, and more. · p6 — Digest's 'labor market flipped from scarcity to surplus' framing restates the source piece's account of the converging pressures breaking the old hiring equilibrium.
explicitRapid change in the shape of engineering roles (driven by AI) is undermining companies' ability to know what to test for even within a traditional interview, independent of the process's pre-existing unreliability.conf: medium
explicitPaid, multi-day 'campfire' work trials on real tickets in a real codebase generate a much stronger hiring signal than a standard interview loop, though weaker than a full internship or co-op.conf: medium
explicitMaking candidate work portable and reusable — a permanent, employer-signed record the candidate keeps regardless of hiring outcome — is what makes unpaid or underpaid trial work fair to candidates and could convert interviewing from a cost center into a profit center.conf: high
extendsPlease stop the AI Confidence Theater · p3 — Elena's essay asserts that verbal interviews no longer work and 'case studies / work trials become necessary'; Yegge extends this into a full mechanism (provisional employment, campfire trials, portable employer-signed credentials) for how such trials could actually work and be made fair. That extraction's gap g3 flags work trials as costly, hard to scale, and gameable — Yegge's 'count work twice' partially addresses the cost/scale half (turning trials into a profit center) but never addresses the gameability half, so it only partially closes that gap (not recorded as a formal closure).
explicitA portable-reputation hiring system only works, and only stays valuable, if employers are honest in what they stamp — inflated or dishonest credentials collapse into the same worthlessness as LinkedIn endorsements.conf: high
implicitAI capability itself (i.e., whether an AI agent can now do the job a candidate is being assessed for) is not the primary driver of the interview's collapse in this argument — the collapse is driven mainly by AI-enabled noise, fraud and a shifting labor-supply/demand balance, on top of a process that was already broken for fifty years before AI existed.conf: medium

Arguments

The traditional interview loop cannot reliably determine whether a candidate can do the job.

Explicit premises:

Google's internal statistical analysis found interviewers routinely disagreed with each other on the same candidate.
Interview scores did not correlate with subsequent on-the-job performance.
In a blind calibration exercise, Google's own Hiring Committee voted to reject about two-thirds of its own members' historical interview packets.

Implicit premises:

A hiring process this internally inconsistent fails a basic test of measurement validity (a valid instrument should not disagree with itself this severely).
The Hiring Committee members were themselves a reasonably representative, successful sample of 'should-have-been-hired' engineers, so their own packets being rejected demonstrates false-negative failure specifically, not just noise.

Inferences:

If interviewer judgment is this unreliable and driven by unconscious bias, individual hiring managers who believe they can 'just tell' during an interview are not meaningfully better calibrated than the group average — undermining any organization's claim to a well-calibrated 'high bar.'

Provisional employment (internships, co-ops) is the best hiring-signal method currently available. strength: low

Explicit premises:

Internship-observed work is a stronger predictor than interview-loop performance.
Geoworks' 6-month co-op (hiring 3 of 8 candidates in 1992) produced what Amazon later called 'by far the strongest group of people they ever brought in.'
University of Waterloo's 6-internship co-op pipeline makes its graduates especially sought-after.

Implicit premises:

Performance during a multi-month provisional role generalizes to performance as a permanent employee (the underlying validity assumption is asserted, not tested against a controlled comparison).
The Geoworks and Waterloo outcomes are attributable to the screening mechanism itself rather than to other confounds (e.g. self-selection of applicants willing to accept underpaid or high-risk arrangements, or Geoworks/Waterloo's independent reputation and applicant pool quality).

Inferences:

If provisional-role performance really does generalize this strongly, standard interviews are not just weak but almost dispensable wherever a provisional-employment stage is available — a stronger claim than 'gold standard' that the author does not explicitly draw.

A portable, employer-signed reputation record attached to real paid work trials ('campfire') can convert interviewing from a cost center into a profit center while being fair to candidates who aren't hired. strength: low

Explicit premises:

Rideshare platforms already mint reusable reputational records (ratings, badges), but they're trapped on-platform and disappear when the worker leaves.
If a candidate's trial work 'counts twice' — usable output plus a portable, permanent, employer-signed record — rejection no longer means walking away with nothing.

Implicit premises:

Employers will resist the incentive to grade-inflate their stamps even though inflation costs them nothing individually and might make them look more generous to attract candidates.
Competing employers will trust and honor stamps issued by other employers without a shared verification standard or governing body.
Candidates, employers, and any portable-credential infrastructure can resolve consent, privacy, and cross-company data-sharing questions without friction.

Inferences:

Because honesty is asserted as the system's entire value basis with no enforcement mechanism, the proposal is structurally a repeat of the 'good faith network' problem that other reputation systems (LinkedIn endorsements, the author's own counterexample) have already failed to solve — the author's own admission ('Candidly, I wanted to hand you a coherent solution today... I decided I'd hold it for now') suggests he recognizes this gap without resolving it.

Narratives

The Google Hiring Committee 'calibration exercise' in which the committee unknowingly reviewed and rejected two-thirds of its own members' original interview packets.
→ Carries the core claim — the interview process is broken — through a single vivid, self-indicting anecdote rather than an argued statistical case; it also implicitly supports the claim that interviewer bias, not signal scarcity alone, is doing real damage.
The Geoworks 6-month co-op (1992), where 3 of 8 candidates were hired, followed years later by Amazon's 1999 acqui-hire of the Geoworks Seattle office and repeated praise from Amazon leadership.
→ Supports the claim that provisional employment is the gold-standard signal, by showing a concrete, memorable case where an unusually demanding provisional screen produced an unusually strong cohort.
Being told by an 'Important Eng Leader' that questioning the interview process was 'farting in church.'
→ Illustrates and supports the 'Change is Taboo' claim that the industry's resistance to reforming interviewing is social/political, not evidence-based — people who passed the gauntlet defend it as a rite of passage.

Questions addressed

✅ Why is the traditional technical interview failing?
It was always statistically unreliable (interviewer disagreement, no correlation with on-the-job performance), and is now additionally strained by AI-enabled résumé/cheating noise, fraudulent applicants, and a possibly loosening labor market that makes provisional roles newly viable for senior candidates.
⭕ What should replace the traditional interview?
Sketched only: paid, multi-day 'campfire' work trials with portable, employer-signed credentials ('count work twice'), explicitly offered as a directional sketch, not a finished solution — the author states he is deliberately withholding a fuller proposal.
✅ How can real-work trials be made fair to candidates who don't get an offer?
By making the trial's output count twice: once as employer signal/usable work, once as a portable, permanent credential the candidate keeps regardless of outcome.
⭕ How would a portable-credential hiring system scale to the volume of hiring that produces today's 'résumé floods,' and how would cross-employer trust in each other's stamps be established and enforced?

Gaps

incompleteThe article never explains how honest employer 'stamping' would be enforced or verified at scale. It asserts a reputational equilibrium ('companies whose stamps mean anything will be the ones known for only handing them out for quality work') without any bootstrapping or policing mechanism — the exact same gap flagged elsewhere in this corpus about work-trial signals under AI-era cheap-talk conditions.
Read next: Spence-style signaling/mechanism-design literature on costly signals and cheap talk; extract-lenny-elena-ai-confidence-theater#g3 in this corpus, which raises the identical unresolved problem from the employer-hiring side.
illogicalDespite the framing (and the prompt's premise) that AI is what's killing the interview, the article's own reliability critique — interviewer disagreement, no correlation with on-the-job performance, the HC calibration story — describes a process broken for fifty years, entirely independent of AI. AI is invoked only as an accelerant (résumé noise, cheating, fraud, labor-supply shock), never as a reason the underlying skills being assessed have changed. The 'AI killed the interview' framing is a rhetorical hook more than an argued causal claim.
Read next: Look for pieces that argue the more specific claim — that AI coding agents change *what* needs to be assessed (e.g. orchestration/judgment vs. from-scratch coding ability) — to see whether that stronger thesis is defended elsewhere with more rigor.
closed byPresentation: Getting Rid of LeetCode Interviews in the World of AI — Argues explicitly for the stronger, more specific thesis that gap flags as missing: that AI changes *what* needs to be assessed in an interview — shifting evaluation from from-scratch algorithmic coding to system-design judgment and the ability to direct/correct an AI collaborator — rather than treating AI merely as an accelerant of pre-existing interview noise and fraud.
incompleteThe provisional-employment prescription assumes candidates can absorb months of underpaid, uncertain trial employment; the article doesn't address the equity dimension — this systematically favors candidates with savings, family financial support, or visa/immigration flexibility, and could disadvantage exactly the population (laid-off senior engineers with mortgages and families, whom the article itself says 'did not want to sign up for a job they might not get to continue') the piece expects to newly accept these deals.
Read next: Literature on unpaid/underpaid internship economies and their exclusionary effects on candidates without financial cushion.
incompleteThe 'campfire' model requires giving as-yet-unvetted candidates paid, multi-day access to a real codebase, real tickets and a real team. The article never discusses the security, IP, confidentiality, or regulatory exposure of that access — a material concern for regulated industries (banking, telecom, healthcare) where even short-term contractor access typically requires background checks, NDAs, and controlled environments that would blunt the model's promised speed and informality.
Read next: Enterprise contractor-access and least-privilege onboarding practices; how regulated industries currently handle short-term/trial engineers.

The Last Technical Interview

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Related documents