t's 2:15 PM on a Thursday and you're sitting across from a candidate in a glass-walled conference room named after a planet or a Muppet. You have forty-five minutes. Your Slack DM from the recruiter says "culture screen." You have a scorecard open in Greenhouse with fields for "communication," "collaboration," and "values alignment." You know, within ninety seconds, whether this person is going to work out. The remaining forty-three minutes are theater — for the candidate, for compliance, for yourself. You cannot say what you know or how you know it. If pressed, you'll say something about "energy" or "how they'd handle ambiguity" or "whether I'd want to pair-program with them." You will believe, sincerely, that you are assessing something real. You are. You just can't say what it is, and that unsayability is not a bug in the process. It is the process.

Organizational psychology has known for at least a decade that "culture fit" interviews reproduce homogeneity. Lauren Rivera's ethnographic work on elite hiring showed that interviewers select for shared leisure activities, conversational styles, and class markers while believing they're evaluating competence. Iris Bohnet's research demonstrated that unstructured interviews are worse than useless — they actively degrade decision quality compared to statistical prediction. The empirical finding is not in dispute. What remains unexplained is a specific and maddening phenomenon: why does adding rubrics, scorecards, and bias training to culture fit interviews fail to change outcomes? Companies pour thousands of hours into structured interview design, calibration sessions, and interviewer certification, yet the people who emerge from the pipeline look, sound, and think remarkably like the people already there. The rubric gets absorbed. The scorecard gets filled out. The same person gets hired. The mechanism of this absorption — the reason reforming the culture fit interview is not merely difficult but structurally impossible — is what theory can explain and data alone cannot.

Wittgenstein's rule-following considerations in the *Philosophical Investigations* are routinely misread as a skeptical claim that rules are meaningless or that anything goes. The actual argument is subtler and, for our purposes, more devastating. In §201, he observes that no course of action is determined by a rule, because every course of action can be *made out to accord* with the rule. But §202 delivers the positive counterpoint: "And hence also 'obeying a rule' is a practice." Rules are real. They govern behavior. People follow them with remarkable consistency. But the normativity of the rule lives in the practice of following it, not in any formulation of it. No finite set of words exhausts what it means to follow the rule, because the practice always outruns its own description.

This is precisely the situation with culture fit criteria. "Strong communicator" is a rule. Interviewers apply it with statistically impressive convergence — panel members agree on who is and isn't a strong communicator at rates well above chance. Something is being communicated; something is being tracked. But try to formalize it. Does "strong communicator" mean speaks clearly? Speaks concisely? Asks good questions? Listens actively? Knows when to stop talking? Can explain a technical concept to a non-technical audience? All of these, sometimes. None of these, reliably. Every attempt to decompose the criterion into sub-criteria simply reproduces the problem at a finer grain. You now need to define "asks good questions," and you're back where you started. The criterion functions not as a description that interviewers match against candidates but as a *shared practice* that interviewers enact together, calibrated not by the scorecard but by the form of life they already inhabit.

This is why the "better rubrics" approach fails — not because rubrics are useless in general, but because layering a rubric onto a culture fit interview is like writing the rules of humor on a whiteboard and expecting them to be funny. The rubric becomes one more element within the language game, interpreted through the same tacit practice it was supposed to constrain. When an interviewer scores a candidate 3/5 on "collaboration," they are not applying the rubric's written definition. They are translating a felt sense — produced by the interaction, by the candidate's body, by forty-five minutes of conversational rhythm — into a number. The rubric provides post-hoc justification for a judgment that was never rubric-derived. It launders intuition into data.

The critical point — and this is where the theoretical apparatus earns its keep against the I/O psychology literature — is that this isn't a failure of discipline or training. It is a constitutive feature of what a culture fit interview *is*. The practice exceeds the formalization not because interviewers are lazy or biased (though they may be both) but because the kind of knowledge being deployed is, in Wittgenstein's sense, knowledge-how that cannot be fully converted into knowledge-that. You can train interviewers to recite criteria. You cannot train them to apply criteria independent of the form of life in which those criteria have meaning. And the form of life in question — the shared habitus of the engineering team, its rhythms of humor and deference and technical argumentation — is exactly the thing that produces homogeneity.

Wendy Chun's concept of "updating to remain the same," developed across *Updating to Remain the Same* and her work on new media's habitual operations, describes how digital networks perpetually refresh their interfaces, protocols, and content in order to maintain underlying structures of power and relation. The new replaces the old constantly — new feeds, new users, new features — yet the network's fundamental logic of capture, sorting, and homophily persists. The *feeling* of novelty is essential to the mechanism. If the system felt static, users would leave. The update is what keeps them inside the same structure.

The disanalogy is obvious: an interview loop is a social institution, not a technical medium. But the analogy is more than metaphorical if we take seriously what Chun means by "habit." For Chun, habit is not mere repetition but the mechanism by which networks naturalize their operations — the point at which the artificial becomes second nature, where the designed vanishes into the obvious. A hiring pipeline is exactly such a habitual system. It runs on cycles. Each cycle ingests new candidates — new résumés, new backgrounds, sometimes conspicuously new demographic profiles — and processes them through the same evaluative apparatus. The output feels fresh. The team announces its new hire with genuine excitement about what this person brings. Yet the dispositional signature of the team remains eerily stable. The same conversational norms. The same hierarchy of what counts as impressive. The same implicit model of what an engineer sounds like when they're being "real."

Chun's framework explains the specific temporal trick that makes culture fit hiring so resistant to critique. At any given moment, the team can point to its most recent hire and say: look, we hired someone with a completely different background. The pipeline *updated*. But the update is precisely what preserved the same underlying sociality, because the interview selected for someone whose differentness was legible within the team's existing evaluative grammar. The new hire is different in the ways the team already knows how to appreciate and same in the ways the team cannot see as dimensions of variation at all. This is not conspiracy. It is habit operating below the threshold of reflective awareness, exactly where Chun locates its power.

This is why diversity initiatives that focus on the top of the funnel — broader sourcing, blind résumé review, diverse candidate slates — produce disappointing results when the culture fit interview remains the bottleneck. The funnel widens, the pool diversifies, and the interview narrows it back to the familiar. The update happens. The same remains.

The interview loop has material affordances that are not neutral. A forty-five-minute one-on-one conversation in a conference room is a specific communicative format that privileges specific capabilities: rapid verbal processing, comfort with sustained eye contact, the ability to perform spontaneous intimacy with a stranger who holds power over your economic future, fluency in the register of casual-but-smart that codes as "senior engineer" in the Bay Area or "solid developer" in the Midwest. These are not universal human capabilities. They are culturally specific, neurologically variable, and class-inflected skills that correlate poorly with the ability to write reliable software, review code generously, or debug a production incident at 2 AM.

The lunch interview is worse. Now you are assessing someone's table manners, their comfort with unstructured social time, whether they order a drink, whether they know how to navigate the micro-politics of who picks up the check. The material setting — a restaurant, a café, a team kitchen — encodes an entire sociology of ease and belonging. Autistic engineers, engineers from working-class backgrounds, engineers whose first language isn't English, engineers who don't drink — all are subtly penalized not for lacking technical skill or collaborative capacity but for failing to perform a specific middle-class sociality that the format demands but never names.

And because the format never names what it demands, the demand cannot be challenged. An interviewer who says "I didn't feel a connection" after a lunch interview is not lying. They genuinely didn't feel a connection. The feeling is real. But the connection they didn't feel is a connection that the material format of the interview produced as the relevant signal — conversational ease in a specific register, in a specific setting, under specific power dynamics. A different format would produce different signals. The medium is not transparent to the message. The medium *is* the message, in the most banal and literal McLuhanian sense: what the interview can transmit is determined by what the interview *is*, and what the interview is selects for a narrow band of human sociality while appearing to select for something universal.

Here is where the theoretical apparatus must outperform the existing literature or earn its retirement. Organizational psychology already tells us to replace unstructured interviews with structured ones. Work-sample tests, standardized behavioral questions with anchored rating scales, blind code reviews — all of these outperform culture fit interviews on every metric that matters: predictive validity, demographic equity, inter-rater reliability. This is not news. Laszlo Bock said it in 2015. The meta-analyses confirmed it long before that.

But the theory predicts something the I/O psychology literature does not, and this is the claim I want to make sharply: *structured values-alignment interviews will also fail to the degree that they retain the conversational form*. This is the actually controversial prediction. Many companies that have "abolished" culture fit interviews have replaced them with something called a "values interview" or "collaboration interview" — structured, yes, with rubrics and calibrated scoring, but still fundamentally a conversation between an interviewer and a candidate, still forty-five minutes in a room, still governed by the tacit evaluative practices of the interviewing team. The theory predicts that these interviews will reproduce homogeneity at rates statistically indistinguishable from the old culture fit screen, because the language game persists even when the vocabulary changes. "Culture fit" becomes "values alignment" becomes "collaboration style" — the words update, the practice remains.

The genuinely structural alternative is not a better conversation but the abolition of the conversation as an evaluative instrument for cultural compatibility. Pair-programming on real work. Trial projects. Structured decision-making exercises with observable outputs. Anything that produces *behavioral evidence* rather than *conversational impression*. The distinction is not between structured and unstructured but between formats where the evaluative signal is the candidate's speech and formats where the evaluative signal is the candidate's work. Only the latter escape the language game, because only the latter remove the interviewer's felt sense of "connection" as a decision-relevant input.

This is a harder sell than "use rubrics." It requires engineering leaders to accept that their intuitions about people — honed over years, feeling exquisitely calibrated — are not instruments of perception but instruments of reproduction. The interview doesn't reveal culture. It *generates* culture, one forty-five-minute conversation at a time, each one a small act of renewal that keeps everything the same.