The candidate is
not the profile

1.Introduction: The Category Mistake in AI-Led Executive Assessment

The use of artificial intelligence in hiring is usually discussed through three familiar lenses: efficiency, fairness, and automation. Efficiency asks whether AI can make recruitment faster. Fairness asks whether AI reduces or reproduces bias. Automation asks whether AI will replace human recruiters, sourcers, interviewers, or assessors. These are important questions, but they do not exhaust the problem raised by AI in senior hiring. The deeper issue is conceptual: what kind of thing is a candidate, and what kind of understanding is required to assess one?

AI systems are increasingly capable of producing convincing candidate representations. They can parse a CV, infer skill clusters, summarize career trajectories, identify adjacent industries, rank profiles against job descriptions, generate interview notes, and compare a candidate's background to historical hiring patterns. These outputs can be useful, especially where the task is to enlarge the field of visibility. But visibility is not yet understanding. A system can make a candidate more searchable without making the candidate more understood. It can retrieve a profile without grasping what that profile means in a concrete organizational context. It can generate a plausible fit narrative without bearing responsibility for the judgment that the narrative invites.

This paper argues that AI-led executive assessment rests on a category mistake when it treats candidate representation as candidate understanding. The term "category mistake" is not used loosely. In Gilbert Ryle's sense, a category mistake occurs when something is treated as belonging to a logical category to which it does not belong (Ryle, 1949; Magidor, 2019). In hiring, the profile and the person do not belong to the same category. A profile is a document, model, or representation. A person is an agent with a history, commitments, motivations, reasons, values, constraints, habits of judgment, and a practical relation to possible futures. The profile can support assessment; it cannot become assessment.

The same mistake appears in a second form: the tendency to treat culture fit as a soft skill. Soft skills are portable capabilities. Communication, conflict management, stakeholder awareness, self-regulation, analytical clarity, and team collaboration can be observed, trained, benchmarked, and compared across contexts. Culture fit is different. It is not an additional skill possessed by the candidate in isolation. It is a relation between a candidate's identity and a specific organizational world. It concerns how values, motivation, decision patterns, leadership style, and operating assumptions meet the actual culture, tempo, governance, politics, maturity, and pressure of a particular company.

The practical consequence is central to KiTalent's methodology. AI can legitimately support market mapping, document screening, deduplication, public-signal organization, and reporting. It should not replace consultant-led interviews, identity assessment, stakeholder calibration, or final executive-search judgment. The reason is not that AI is useless. On the contrary, AI is valuable precisely because it can perform representational work at scale. The problem is that representational competence can be mistaken for comprehension.

The article therefore defends a disciplined boundary. It is not a defense of unstructured human intuition against algorithms. That would be both empirically weak and philosophically lazy. Selection research repeatedly shows that unstructured human judgment is noisy and often inferior to mechanical data combination (Meehl, 1954; Grove et al., 2000; Kuncel et al., 2013; Highhouse, 2008). The alternative to AI-led assessment is not gut feeling. It is structured, evidence-based, consultant-led judgment supported by AI where AI is epistemically appropriate.

The paper develops this boundary through seven analytic moves. First, it diagnoses profile/person substitution as a category mistake. Second, it uses Dennett's "competence without comprehension" to explain why AI may be powerful in mapping without understanding the candidate. Third, it draws on Searle, Harnad, and Bender and Koller to distinguish syntactic or linguistic form from grounded meaning. Fourth, it uses Wittgenstein to show that hiring vocabulary is not self-interpreting; terms such as "leadership," "ownership," and "culture fit" acquire their meaning inside practices. Fifth, it uses Sellars and Brandom to argue that a senior interview is not data extraction but a normative practice of giving and asking for reasons. Sixth, it draws on Polanyi, Suchman, and Schön to explain tacit and situated judgment. Seventh, it integrates organizational psychology to distinguish capability, soft skills, identity, and culture fit.

The business applications are then direct. AI belongs in mapping because mapping is representational. AI does not replace identity assessment because identity-fit is relational, dialogic, normative, and situated. Blind CVs are weak evidence because they are decontextualized representations without epistemic accountability. Proof-First Search is stronger because it shifts proof from promise and profile-theatre to situated evidence: real candidates, real employers, real compensation expectations, real motivation, real availability, real mobility constraints, and real assessment against a real mandate. Speed has qualitative value because senior talent availability is not static; it is temporal and perishable.

The central claim can be stated simply: mapping is not judgment; representation is not assessment; a profile is not a person.

2.The Candidate Is Not the Profile: Category Mistakes and Executive Search

Ryle introduced the notion of category mistake to expose a specific form of conceptual confusion. His famous examples are well known. A visitor is shown the colleges, libraries, laboratories, offices, and lecture halls of a university and then asks, "But where is the University?" The visitor has not missed a building; he has misunderstood the category to which "university" belongs. The university is not an additional object alongside the buildings. It is the institutional organization of those buildings and practices. In another example, a spectator sees the players in a cricket match and then asks which player is responsible for the team spirit. The spectator has again misunderstood the category. Team spirit is not a separate action alongside batting, bowling, and fielding. It is a manner in which the team acts.

These examples are unusually useful for executive search. A candidate profile is like the list of buildings. It gives access to visible components: job titles, companies, dates, education, keywords, reported achievements, sectors, geographies, and sometimes compensation data. These components matter. Without them, search would be blind. But the candidate is not the sum of profile components in the same way that the university is not an additional building. The person is a living integration of history, capability, self-understanding, motivation, values, risk tolerance, decision style, relationships, constraints, and future orientation.

Treating the profile as the person is therefore not merely an empirical error. It is not simply that the profile is incomplete and needs a few more data points. The profile belongs to a different logical type. It is a representation of the candidate, not the candidate as an assessable person. Adding more representation does not automatically cross the category boundary. A longer CV, a richer LinkedIn page, an AI-generated profile summary, a skill graph, a personality proxy, and a ranking score can all make the representation more elaborate while leaving the category mistake intact.

The same structure applies to culture fit. If team spirit is not a separate cricket role, culture fit is not a separate soft skill. Soft skills are capabilities. They concern what a person can do or how they can behave across a range of contexts: communicate clearly, manage conflict, listen actively, influence stakeholders, structure a conversation, lead a meeting, or handle ambiguity. Culture fit is a relation. It concerns whether a particular person's values, motivations, identity patterns, decision habits, and operating style can cohere with a specific organizational environment.

This difference is central to Montaruli's capability-identity distinction developed in KiTalent's research. Capability asks what a person can do. Identity asks who the person is in relation to the role, the organization, the stakes, and the future implied by the mandate. A senior candidate may have the required capabilities and still fail identity-fit. They may be technically excellent and politically misaligned. They may communicate well but resist the governance model. They may have led transformation in one environment but lack the motivation to live through another. They may possess the soft skill of stakeholder management but reject the stakeholder structure they would inherit.

The category mistake becomes especially dangerous when AI systems are marketed as fit engines. A fit score can be useful as a prompt for inquiry, but it can easily conceal the fact that different kinds of fit are being collapsed. Person-job fit and person-organization fit are distinct constructs in organizational psychology (Kristof-Brown, Zimmerman, & Johnson, 2005). A system that detects job-skill similarity may be estimating one aspect of person-job fit. It is not thereby assessing person-organization fit, values congruence, motivation, leadership identity, or the candidate's relation to a specific mandate.

An AI tool can compare representations. It can say, for example, that a candidate's career pattern resembles previously successful executives. But resemblance is not fit. Similarity is not judgment. Ranking is not understanding. Executive assessment asks a different question: does this person, with this history, motivation, identity, and operating style, make sense for this organization now? That question cannot be answered by treating the candidate as if they were exhausted by their profile.

3.Competence Without Candidate Understanding

Daniel Dennett's phrase "competence without comprehension" provides a useful way to avoid both naive anti-AI and naive AI maximalism. Dennett uses the phrase to describe how complex functional performance can arise without reflective understanding. A system may competently execute a task without comprehending the reasons, meanings, or purposes that a human observer associates with the task (Dennett, 2017). This distinction is particularly relevant to AI in hiring because many AI tools are genuinely competent. The point is not to deny their competence. The point is to locate it correctly.

In executive search, AI can exhibit mapping competence. It can process large candidate universes, search across public professional data, identify keywords and adjacent functions, summarize profiles, organize career histories, enrich company lists, de-duplicate records, and structure reports. These are not trivial tasks. They can substantially improve the efficiency and coverage of search. In a fragmented talent market, where relevant candidates may be dispersed across functions, geographies, titles, and industries, this representational competence can give consultants a wider and cleaner field from which to work.

But mapping competence is not candidate comprehension. To understand a candidate in the executive-search sense is not merely to process facts about that candidate. It is to interpret the person's career trajectory, capability, identity, motivation, values, timing, constraints, and operating style in relation to a concrete mandate. Candidate understanding is not a mental state attributed to the consultant as a mystery. It is an achievement within a structured professional practice: evidence is gathered, claims are tested, narratives are probed, reasons are examined, inconsistencies are explored, motivations are calibrated, and the candidate is interpreted against the client's actual context.

Dennett must be used carefully here. He is not a simple anti-AI thinker. His broader philosophy is explicitly naturalistic and evolutionary, and he does not rely on a magical division between human understanding and machine performance. His framework is useful precisely because it allows a non-romantic distinction. AI competence can be real. It can be economically valuable. It can even outperform humans in defined tasks. But competence in one domain does not entail comprehension in another.

The mistake in AI-led executive assessment is the inference from successful profile processing to candidate understanding. A system that can summarize a candidate's career does not thereby understand the candidate's motives. A model that can infer that a candidate is technically aligned does not thereby grasp whether the candidate can live inside the politics, pace, and value structure of the client's organization. A tool that can generate interview questions does not thereby participate in the normative exchange by which a candidate owns, revises, or defends their reasons.

The correct conclusion is therefore not "AI cannot do search." It is the opposite: AI can do important parts of search because those parts are representational. The conclusion is that AI should not be confused with the part of search that requires understanding. AI has mapping competence. It does not have candidate comprehension.

4.Syntax, Form, and Grounded Meaning

The distinction between representation and understanding has a long history in philosophy of mind and language. John Searle's Chinese Room argument remains the classical reference point. Searle argues that a system could manipulate symbols according to formal rules and produce outputs indistinguishable from genuine linguistic competence without understanding the meanings of those symbols (Searle, 1980; Cole, 2024). The purpose of the argument is not to deny that computers can process language-like structures. It is to show that syntactic manipulation is not sufficient for semantic understanding.

Modern AI systems differ from the rule-based systems that Searle originally had in mind. Large language models are not simply hand-coded symbol manipulators. They learn statistical relationships from enormous datasets and can produce highly flexible outputs. This matters, and a serious paper should not pretend that contemporary AI is merely a 1980s expert system with more memory. Still, Searle's distinction remains useful: the formal or statistical manipulation of language does not by itself establish grounded understanding.

Harnad's symbol-grounding problem sharpens the issue. A formal symbol system manipulates tokens. But how do those tokens acquire meanings that are intrinsic to the system rather than parasitic on meanings supplied by human interpreters (Harnad, 1990)? Bender and Koller make a related point in contemporary computational linguistics. They distinguish linguistic form from meaning and argue that systems trained only on form cannot, by that training alone, acquire human-analogous natural language understanding, because understanding involves communicative intent and grounding in the world (Bender & Koller, 2020). Bender et al.'s "stochastic parrots" critique further warns against equating plausible language generation with understanding, while also emphasizing the social and ethical risks of language models trained on large-scale text distributions (Bender et al., 2021).

The application to executive search is direct. A candidate profile is a linguistic artifact. It contains symbols: titles, verbs, company names, dates, claims, achievements, metrics, and self-descriptions. An AI system can process these symbols extremely well. It can infer patterns and generate useful summaries. But the meaning of a candidate's career cannot be grounded in profile syntax alone.

Consider the sentence: "Led transformation across multiple geographies." As a linguistic form, it is clear enough. As assessment evidence, it is almost empty until grounded. What kind of transformation? Operational, cultural, commercial, technological, financial, regulatory, or post-merger? What counted as success? Who resisted? What authority did the candidate actually hold? Was the transformation designed by the candidate or inherited? What was the governance structure? What was sacrificed? What was learned? Why did the candidate leave? Would they choose that kind of mandate again?

The same phrase can mean very different things in a founder-led scale-up, a regulated bank, a private-equity portfolio company, a state-linked enterprise, a family business, or a multinational matrix. The profile provides linguistic form. Assessment requires grounded inquiry.

This is not an argument that AI-generated candidate summaries are useless. They can be useful precisely as summaries. They can help prepare the consultant, structure a search process, highlight gaps, or make a profile easier to read. But they should not be treated as grounded assessment evidence. The candidate's meaning in the hiring process emerges when the profile is tested against reality: employer context, mandate context, compensation expectations, mobility, current motivation, constraints, reasons for interest, and the candidate's own account of their decisions. Until that happens, the profile remains representation.

5.Meaning as Use: Hiring Language-Games

Wittgenstein's later philosophy gives a second way to understand why hiring language is not self-interpreting. In the Philosophical Investigations, Wittgenstein moves away from the idea that words have meaning by standing for objects in a simple representational scheme. Meaning is tied to use, and use is embedded in language-games: socially organized practices in which words function as moves (Wittgenstein, 1953/2009; Biletzki & Matar, 2024). A word does not carry one fixed essence that can be extracted independently of its practical setting. It acquires its force through the role it plays in a form of life.

Hiring language is full of terms that appear transparent but are actually practice-dependent. "Leadership" does not mean the same thing in a 20-person product company and a 40,000-person regulated multinational. "Ownership" does not mean the same thing in a founder-led culture and a governance-heavy public company. "Stakeholder management" does not mean the same thing where the stakeholder is a family owner, a private-equity sponsor, a board audit committee, a union, a regulator, or a global matrix. "Entrepreneurial" can mean commercially aggressive, institutionally immature, resourceful, under-controlled, or allergic to process depending on context.

The term "culture fit" is especially vulnerable. Without a disciplined language-game, it can become a placeholder for chemistry, similarity, taste, class markers, or implicit bias. Rivera's study of elite professional service firms shows how cultural matching can influence hiring in ways that go beyond productivity and competence, often privileging shared lifestyle, self-presentation, and cultural similarity (Rivera, 2012). This is precisely why culture fit must be carefully defined. It should not mean "people like us." It should mean identity-context coherence under the actual conditions of the mandate.

Wittgenstein helps explain the weakness of both blind CVs and AI-generated candidate summaries. They present language detached from the hiring practice that would give it assessment meaning. A blind CV may say "excellent strategic leadership." An AI summary may say "strong culture fit for dynamic organizations." But unless those phrases are embedded in a concrete practice of assessment, they remain underdetermined. What counts as strategic? What counts as dynamic? What organizational culture is being invoked? What evidence supports the claim? Which parts of the candidate's story confirm it, and which parts complicate it?

A consultant-led process does not simply collect words; it reconstructs their use. It asks what the candidate meant by a claim, how the claim functioned in the candidate's prior organization, whether the claim can be corroborated, and whether the same behavior would have the same meaning in the client's world. Assessment is therefore not simply semantic matching. It is pragmatic interpretation.

This point also supports the distinction between soft skills and culture fit. A soft skill can often be operationalized as a repeatable capability. Culture fit cannot be reduced to a label because its meaning depends on the organizational language-game. What counts as "direct communication," "high ownership," "collaborative leadership," or "low-ego execution" varies across contexts. Therefore, the assessor must interpret use, not merely parse form.

6.The Space of Reasons: Why Interviews Are Not Data Extraction

If an executive interview is not simply an event in which information is extracted from a candidate, what is it? Sellars and Brandom provide the conceptual vocabulary. Sellars argues that to characterize an episode as knowledge is not merely to describe it empirically but to place it in the logical space of reasons: the space of justifying and being able to justify what one says (Sellars, 1956; deVries, 2024). Brandom develops this pragmatist insight into an account of discursive practice as a social game of giving and asking for reasons. To make a claim is not merely to emit a sentence. It is to undertake a commitment that can be challenged, defended, revised, and connected inferentially to other commitments (Brandom, 1994).

This framework is unusually powerful for executive assessment. A senior interview is not merely a data collection session. It is a reason-giving practice. The candidate does not only state facts. They explain decisions, justify trade-offs, contextualize failures, defend priorities, revise claims when challenged, and reveal what they treat as obvious, negotiable, protected, risky, or meaningful. The consultant does not merely record answers. The consultant tracks commitments: what the candidate claims to have done, what they take responsibility for, what they avoid, what they can explain, where the narrative is coherent, where it is evasive, and how the candidate's reasons relate to the mandate.

Brandom's language of deontic scorekeeping can sound technical, but the underlying idea is practical. In a conversation, participants keep track of what others are committed to and what they are entitled to claim. In executive search, a consultant performs a professional version of this practice. If a candidate claims to be a transformation leader, what follows? What evidence are they committed to producing? What failures must they be able to explain? What trade-offs should they understand? What reasons can they give for wanting a similar or different mandate now? What does their account entitle the consultant to infer, and what remains unearned?

AI can simulate parts of this exchange. It can generate interview questions, transcribe answers, summarize themes, detect keywords, and identify inconsistencies in text. These are useful supports. But the normative structure of the interview is not exhausted by text processing. The candidate is accountable to another person in a context of trust, discretion, persuasion, and mutual interpretation. A senior candidate's answer matters because it is owned. It can be challenged. It can be withdrawn. It can affect reputation. It can alter the candidate's relation to the opportunity.

The phrase "candidate understanding" should therefore be defined carefully. It is not an ineffable mental intuition. It is the structured interpretation of a candidate's capability, identity, motivation, values, timing, and operating style through a process in which the candidate gives reasons and the consultant evaluates those reasons against the mandate. Candidate understanding is achieved in the space of reasons, not merely in the space of data.

This explains why AI should not conduct or replace senior identity assessment. AI can help prepare the interview. It can help document it. It can help organize evidence after it. But it does not bear the interpersonal and normative role through which a candidate is asked to own reasons in relation to a real professional decision.

7.Tacit Knowledge and Situated Assessment

A further reason why candidate understanding cannot be reduced to profile data lies in the tacit and situated character of expert judgment. Polanyi's classic claim that "we can know more than we can tell" is not an excuse for vagueness. It is a claim about the structure of human expertise (Polanyi, 1966). Experts often rely on pattern recognition, contextual sensitivity, embodied familiarity, and interpretive discrimination that cannot be fully articulated as explicit rules. A person can recognize a face, a tone shift, a weak argument, a false confidence, or a credible hesitation without being able to reduce the recognition to a complete checklist.

In executive search, tacit judgment appears in many ways: how a candidate narrates failure, how they respond to challenge, which parts of their story become over-polished, where the energy rises or falls, what they consider a serious problem, whether they understand the political structure of a mandate, whether they over-index on title, how they talk about former teams, what they protect, and what kind of risk they seem to seek or avoid. These are not mystical signals. They are professional evidence, but often evidence that emerges through situated interaction rather than pre-coded data.

Suchman's critique of classical AI reinforces the point. In Plans and Situated Actions, she argues that human action is not merely the execution of a prior plan. Plans are resources for situated action, but action unfolds in relation to changing circumstances, social interpretation, and practical contingency (Suchman, 1987). Schön's account of the reflective practitioner similarly emphasizes reflection-in-action: professionals do not simply apply rules; they think through situations while acting within them (Schön, 1983).

The consultant's expertise should be understood in this way. It is not unstructured intuition. It is structured practical expertise. A consultant begins with a calibrated mandate, builds a market map, engages candidates, uses interview frameworks, records evidence, and compares candidate patterns against role requirements. But the assessment also evolves as the conversation unfolds. A candidate's unexpected answer may change the line of inquiry. A contradiction may reveal a deeper issue. A motivation that appeared weak may become compelling once constraints are understood. A strong profile may weaken when the candidate cannot explain the actual stakes of the mandate.

AI can support this practice but not replace it. It can store evidence, propose questions, summarize transcripts, and compare stated skills to the brief. It cannot inhabit the situated professional exchange in which identity-fit emerges. The candidate's identity is not fully contained in the explicit data. It appears through the way explicit claims are made, defended, qualified, and related to context.

The risk, again, is category confusion. If one believes that all relevant assessment evidence is explicit, codifiable, and already present in the profile, AI assessment appears plausible. If one understands senior assessment as a situated practice involving explicit evidence, tacit judgment, and normative reason-giving, then AI's proper role becomes clear: it is a support for evidence organization, not the owner of judgment.

8.Capability and Identity: Why Culture Fit Is Not a Soft Skill

The analytic distinctions developed so far must be translated into organizational psychology. The capability-identity distinction is not a poetic contrast. It corresponds to different evidentiary registers.

Capability concerns what a person can do. It includes technical skills, management skills, leadership behaviors, and many soft skills. Capability can often be evidenced through track record, performance examples, structured interviews, work simulations, references, psychometric tools, and comparison with role demands. It is not always easy to assess, but it is relatively more representable.

Identity concerns how the person understands and inhabits their professional self in relation to values, motivation, responsibility, power, risk, recognition, conflict, growth, and organizational context. Identity is not a fixed essence hidden behind behavior. It is a pattern of commitment and self-interpretation that becomes visible through narrative, decisions, trade-offs, and the reasons a person gives for what matters to them. McAdams's work on narrative identity is relevant here: people make sense of their lives through stories that organize past, present, and future (McAdams, 2001). Leader identity research similarly distinguishes leader competencies from the deeper internalization of oneself as a leader (Kragt & Day, 2020).

Person-environment fit research supports this distinction. Kristof-Brown, Zimmerman, and Johnson's meta-analysis distinguishes person-job fit, person-organization fit, person-group fit, and person-supervisor fit (Kristof-Brown et al., 2005). Person-job fit is closer to capability: do the candidate's abilities, experience, and preferences align with task demands? Person-organization fit is closer to identity: do the person's values, needs, and characteristics align with the organization's culture and norms?

Soft skills belong primarily to capability. A candidate may have strong communication, emotional regulation, team collaboration, and stakeholder management. These skills matter. They can be observed and assessed. But culture fit is not identical with the possession of these skills. Culture fit asks whether the candidate's identity and operating style will cohere with the organization's actual environment.

This is why the phrase "culture fit and soft skills are not the same" is conceptually important. Treating culture fit as a soft skill makes it seem portable, isolatable, and scoreable. But culture fit is relational. It is not located inside the candidate in the same way as a communication skill. It appears in the relation between candidate and organization.

This does not mean culture fit should be vague. On the contrary, because culture fit can become a vehicle for bias, it must be more disciplined than ordinary skill assessment. Rivera's work shows how culture fit can degrade into cultural matching and homophily (Rivera, 2012). The response is not to abandon culture fit or to automate it through opaque tools. The response is to define it more rigorously: values alignment, decision style, governance tolerance, pace, conflict posture, stakeholder logic, risk orientation, and motivation for the actual role.

Thus, the remedy for biased culture fit is not automated opacity. It is structured, explicit, consultant-led identity assessment. AI can help organize evidence about capability. It can even help detect themes relevant to identity. But it should not be treated as the assessor of identity-fit, because identity-fit is a relational judgment about a person and a context.

9.The Algorithmic Counterargument: Structured Judgment, Not Intuition

Any argument for consultant-led judgment must confront the strongest objection: algorithms often outperform unaided human judgment. Meehl's work on clinical versus statistical prediction initiated a long tradition showing that mechanical methods of combining data can equal or outperform informal human judgment across many domains (Meehl, 1954). Grove et al.'s meta-analysis found mechanical predictions of human behavior to be equal or superior to clinical prediction across a wide range of circumstances (Grove et al., 2000). Kuncel et al. extended this logic to selection and admissions decisions, finding that mechanical data combination often improves predictive validity compared to holistic combination by human decision-makers (Kuncel et al., 2013). Highhouse criticized the stubborn reliance on intuition and subjectivity in employee selection (Highhouse, 2008).

This evidence should be conceded, not minimized. Human interviewers are subject to bias, noise, overconfidence, halo effects, similarity attraction, narrative seduction, and premature closure. Unstructured interviews can be unreliable. Cultural fit can be abused. A consultant's intuition can be wrong.

The conclusion, however, is not that AI should own identity assessment. The conclusion is that human judgment must be structured. Kahneman and Klein's analysis of intuitive expertise is helpful: intuitive judgment is more likely to be valid where the environment is sufficiently regular and where the expert receives timely, high-quality feedback (Kahneman & Klein, 2009). Executive hiring is often a low-feedback, high-complexity environment. The consequences of a mis-hire may not become clear for months. The mandate may be strategically unique. Comparable data may be limited. The environment may be politically complex and rapidly changing.

This makes unstructured intuition dangerous. But it also makes fully automated identity judgment unstable. If the assessment object is not a stable, repeatable, well-specified target, then historical pattern matching can create false confidence. AI may be strong in organizing evidence and weak in judging what that evidence means in a unique mandate.

The correct boundary is therefore neither human intuition nor algorithmic replacement. It is structured consultant judgment supported by AI. This includes calibrated briefs, explicit fit criteria, separation of capability and identity, structured interviews, documented evidence, market mapping, reference triangulation, stakeholder calibration, and disciplined challenge to first impressions. AI can improve the inputs. It can reduce search friction. It can organize candidate information. It can make comparisons more systematic. But it should not be asked to decide the identity-context relation.

The most important sentence of this paper may therefore be this: the alternative to automated assessment is not intuition; it is structured judgment.

10.Why AI Belongs in Mapping but Not Identity Assessment

The preceding sections justify a practical boundary. AI belongs in mapping because mapping is representational. Identity assessment remains consultant-led because identity-fit is grounded, relational, normative, tacit, and situated.

Mapping asks questions such as: who exists in the market? Which companies are adjacent? Which titles may hide relevant functions? Which candidates have comparable experience? Which profiles are duplicates? Which public signals suggest recent movement, promotion, or role expansion? Which geographies or industries should be included? Which technical capabilities appear in the visible field?

These questions are well suited to AI-assisted workflows. AI can expand coverage, accelerate discovery, reduce repetitive document work, improve pipeline organization, and help consultants prepare more intelligently. In KiTalent's methodology, AI is positioned in this data-intensive zone: talent mapping, document screening, deduplication, market signals, and reporting. That use is not merely defensible; it is desirable. A consultant should not waste senior judgment on tasks that software can perform more quickly and consistently.

Identity assessment asks different questions: why this role, now? What does the candidate want to protect or change? Which forms of authority do they accept or resist? How do they interpret failure? What kinds of conflict do they avoid? What motivates them beyond compensation and title? Can they operate inside this stakeholder structure? Are their values compatible with the organization's real decision norms? Are they running from a current situation or moving toward this mandate? What would cause them to fail after twelve or eighteen months?

These questions are not reducible to profile comparison. They require dialogue, interpretation, evidence, judgment, and accountability. An AI system can suggest possible questions or summarize the candidate's responses. It cannot assume responsibility for the meaning of the answers. It cannot engage in the same trust relation. It cannot know the client world through lived professional contact. It cannot own the consequences of a wrong interpretation.

This boundary is increasingly aligned with regulatory caution. The EU AI Act classifies AI systems used for recruitment or selection, including filtering applications and evaluating candidates, among high-risk systems in employment contexts (European Parliament and Council, 2024). The legal point is not a philosophical proof, but it reinforces the normative direction: employment AI is high-stakes, and meaningful human oversight is not optional window-dressing.

The point is not that a human must rubber-stamp an AI recommendation. Rubber-stamp oversight is not judgment. Meaningful oversight requires that consultants understand the mandate, interrogate the evidence, challenge the output, engage the candidate, and remain accountable for the shortlist. AI should widen the field of visibility. Consultants must remain responsible for the meaning of what becomes visible.

11.Blind CVs as Weak Epistemic Signals

The same analytic framework explains why blind CVs are weak evidence in executive search. This claim must be carefully distinguished from a different and valid practice: anonymized recruiting can reduce bias in certain early-stage, high-volume hiring contexts (Bohnet, van Geen, & Bazerman, 2016). The critique here is not directed against anonymization as a bias-reduction tool. It is directed against blind CV selling in executive search, where anonymized candidate summaries are used as commercial proof that a search firm can deliver a mandate.

A blind CV is a representation stripped of verifying context. It may show experience, claimed achievements, industries, and functional signals. But it does not prove candidate identity, current employer reality, interest in the mandate, compensation expectations, notice period, relocation readiness, motivation, relationship with the consultant, or assessment against the real client context. It does not show that the candidate has been engaged under the conditions that matter. It is a document, not situated evidence.

Spence's signaling theory is helpful here. In markets characterized by information asymmetry, signals matter because one party knows more about underlying quality than another (Spence, 1973). But signals differ in credibility. A strong signal is costly, observable, and difficult to fake. A weak signal is cheap and ambiguous. In executive search, a blind CV is a weak signal because it is easy to produce and difficult for the buyer to verify. In the age of generative AI, it can become weaker still: a plausible candidate representation can be manufactured, enriched, or polished without proving that the candidate is real, available, interested, or assessed.

From the standpoint of symbol grounding, the blind CV is also weak. Its claims are not yet grounded in a real interaction. It is not clear who owns the statements, whether the statements are current, what the candidate meant by them, or how they relate to the mandate. From the standpoint of Wittgenstein, the blind CV removes the language from the practice that would give it hiring meaning. From the standpoint of the space of reasons, the candidate has not yet been asked to give reasons. From the standpoint of category analysis, the blind CV confuses representation with proof.

The problem is not anonymity itself. The problem is decontextualized commercial proof. A blind CV can show that a firm can produce a document. It does not show that the firm can enter the real market, engage relevant candidates, test motivation, interpret identity, and validate a shortlist. In senior search, the proof unit should not be the blind profile. It should be the validated candidate relationship.

This is why KiTalent's refusal to send blind CVs is not merely a sales preference. It follows from an epistemic principle: proof in executive search must be situated.

12.Proof-First Search as Situated Evidence and De-risking

Proof-First Search can be understood as a response to information asymmetry in professional services. Executive search is a high-trust, high-uncertainty service. The buyer often cannot know in advance whether the search firm will reach the right market, engage the right candidates, interpret the mandate accurately, and produce a relevant shortlist. Traditional retained search asks the client to commit significant fees before seeing evidence of delivery. Blind CV models offer a representation of possible delivery without necessarily proving candidate reality.

Economists use the concept of credence goods to describe services whose quality is difficult for buyers to evaluate, sometimes even after consumption (Darby & Karni, 1973). Executive search has credence-good features. The client may not know whether the search firm mapped the market well, whether stronger candidates were missed, whether candidates were truly interested, or whether the shortlist reflects deep assessment or convenient availability.

Proof-First Search shifts the evidentiary burden. Instead of asking the client to rely primarily on promise, brand, or upfront trust, the firm demonstrates its work through a validated shortlist before the major fee commitment. The evidence is situated: named candidates, current employers, compensation expectations, motivation, mobility, availability, and interview reports against the actual mandate. The proof is not that a plausible representation exists. The proof is that real candidates, under real constraints, have been engaged, assessed, and deemed worth interviewing.

This model can be read through signaling theory. A validated shortlist is a stronger signal than a blind CV because it is costlier to produce and more difficult to fake. It requires actual market entry, outreach, conversation, motivation testing, expectation calibration, and consultant judgment. It therefore reduces information asymmetry more effectively than either an upfront promise or a decontextualized profile.

Proof-First also aligns with the philosophical distinction between representation and understanding. The search firm does not merely present representations; it presents evidence that a structured process has connected candidate representations to candidate reality. This matters because the client is not buying documents. The client is buying a reduction of hiring risk.

The de-risking logic is especially important for senior roles. A failed executive hire is costly not only because of compensation or replacement fees. It can damage strategy, morale, investor confidence, customer relationships, compliance, and leadership credibility. De-risking therefore requires more than volume and speed. It requires situated evidence before commitment.

In this sense, Proof-First Search is not only a commercial model. It is an epistemology of search. It shifts trust from promise to proof, and from representation to situated evidence.

13.Speed as Qualitative Access to Time-Bound Candidate Possibilities

Speed is often misunderstood as a purely operational metric. In routine hiring, speed may be measured by time-to-fill or time-to-hire. In senior executive search, speed has a deeper qualitative function: it preserves access to time-bound candidate possibilities.

A senior candidate market is not static. Passive candidates may be open for a short period because of promotion cycles, strategic uncertainty, family timing, relocation windows, compensation events, board changes, funding rounds, burnout, post-acquisition ambiguity, or competing offers. A person reachable today may be unreachable three weeks later. A candidate interested before a bonus cycle may become unavailable after it. A leader open during a moment of strategic frustration may recommit once the current employer changes their scope or incentives.

In this environment, time does not merely measure the process. Time changes the candidate field. The relevant market is not a database of stable records. It is a changing set of people whose availability, interest, constraints, and motivation shift over time.

This is one of the strongest arguments for AI-assisted mapping. AI can accelerate the representational phases of search, allowing consultants to reach a broader and cleaner field sooner. It can reduce manual friction in identifying relevant companies, adjacent titles, market signals, and candidate histories. It can help preserve optionality by preventing the search from becoming narrow or stale before engagement begins.

But the qualitative value of speed depends on the boundary defended in this paper. Speed should not mean rushed judgment. It should mean faster access to the field so that disciplined judgment can be applied while the field is still alive. AI can accelerate mapping; it should not compress identity assessment into a score.

The correct formula is therefore: fast mapping, structured judgment. Speed is not the enemy of quality when it protects access to candidates before the window closes. Speed becomes the enemy of quality only when it substitutes for assessment.

This is why the qualitative edge of speed in talent mapping is not a contradiction. It is a methodological principle. AI increases the pace of visibility. Consultants preserve the quality of meaning.

14.Conclusion: AI-Assisted Judgment, Not Automated Judgment

This paper has argued that AI-led executive assessment becomes conceptually unstable when it treats candidate representation as candidate understanding. The candidate is not the profile. A profile can support judgment, but it is not the object of judgment itself. The object is a situated person in relation to a concrete mandate.

The analytic argument proceeds through several converging distinctions. Ryle shows that profile/person substitution is a category mistake. Dennett shows that competence does not entail comprehension. Searle, Harnad, and Bender and Koller show that linguistic form and symbol manipulation do not equal grounded meaning. Wittgenstein shows that hiring language is not self-interpreting but depends on use within practices. Sellars and Brandom show that understanding involves the normative space of reasons. Polanyi, Suchman, and Schön show that expert judgment is tacit, situated, and practical. Organizational psychology shows that person-job fit and person-organization fit are distinct, and that soft skills and culture fit belong to different evidentiary registers.

The implications for executive search are practical.

First, AI should be used where the task is representational: mapping, screening, deduplication, signal organization, reporting, and preparation. These are legitimate and valuable uses. They make the talent market more visible.

Second, AI should not replace consultant-led identity assessment. Identity-fit requires grounded interpretation of values, motivation, narrative, decision patterns, timing, and operating style in relation to a concrete organizational context. That judgment occurs through structured dialogue, evidence, and responsibility.

Third, culture fit is not a soft skill. Soft skills are portable capabilities. Culture fit is a relational identity-context judgment. Because it is vulnerable to bias, it should be made more structured, not more opaque.

Fourth, blind CVs are weak evidence in executive search. They are representations without sufficient epistemic accountability. They may have a role in bias-reduction contexts, but as commercial proof of search capability in senior mandates they are inadequate.

Fifth, Proof-First Search provides a stronger de-risking model because it replaces promise with situated evidence. It shows that real candidates have been reached, engaged, and assessed against a real mandate.

Finally, speed has qualitative value. In senior hiring, candidate availability is perishable. AI-assisted mapping can preserve access to time-bound possibilities, provided it does not replace the slower and more demanding work of judgment.

The position advanced by Alessio Montaruli through KiTalent's methodology is therefore not anti-AI and not anti-algorithm. It is a defense of the proper division of epistemic labor. AI should organize the field. Consultants must judge the person. Technology can extend market intelligence, but it cannot assume responsibility for the meaning of what becomes visible.