The Psychology of People Who Only Want Clean Work: Tolerance for Ambiguity, Need for Cognitive Closure, and Survival in the AI Era

• Target audience: People who insist on spec finalization, people frustrated by those who do, knowledge workers worried about AI-era careers
• Prerequisites: None
• Reading time: 14 minutes
• How this article relates to the series: This is the research-based detailed reference article. If you want practical tactics first, start with the playbooks below based on your type.
  • The Spec-Driven Engineer’s AI-Era Playbook (for spec-driven type)
  • The Explorer’s AI-Era Playbook (for exploration-driven type)
  • How Instruction-Waiting Workers Can Survive the AI Era (for instruction-waiting type and managers)

Overview

“I want to nail down the specs and requirements, then build exactly to them.” There’s always someone on a team who insists on this. And there’s always someone else who says, “Let’s just get something running and fix it as we go.” Same skill level, same role, same tenure — yet the gap between these two styles is shockingly consistent.

This isn’t a matter of laziness versus diligence. It isn’t raw intelligence either. Psychologically, it can be explained as a difference in how someone processes uncertainty — a cognitive trait independent of IQ. Specifically, this is the domain of Tolerance for Ambiguity and Need for Cognitive Closure, psychological constructs with nearly 80 years of research behind them¹²³.

But AI has suddenly changed the landscape. The cost of “just build it and see” has dropped dramatically over the past decade. Andrej Karpathy’s early-2025 proposal of “vibe coding” reached the point where 25% of Y Combinator’s 2025 Winter batch had codebases that were 95% AI-generated⁴. The logic of “nail down every detail before starting” is structurally wobbling.

A natural question arises: Can “people who can only work by the rules” win in the AI era?

The short answer: “They can’t win” is too quick. But “nothing changes” is also wrong. This article analyzes the “only-wants-clean-work” phenomenon through three psychological constructs, uses March (1991)’s exploration/exploitation framework⁵ and Dell’Acqua et al. (2023)’s “jagged frontier”⁶ as auxiliary lenses, and verifies how this trait operates in the AI era. Practical tactics are delegated to the corresponding playbooks, while this article focuses on “why we can say this” for readers who want the research-based explanation.

1. The Observed Phenomenon — Two Work Styles

“Spec-Driven” and “Exploration-Driven”

In software development environments, two clearly distinct work styles exist.

Spec-driven

• Wants detailed requirements before starting
• Feels anxious with vague instructions
• Wants to eliminate “what about edge cases?” first
• Wants design review before writing
• Strong resistance to “just build it”

Exploration-driven

• Wants to build something that runs and feel it out
• Can move with direction, not full specs
• Handles edge cases when they appear
• Discovers design while writing
• Strong resistance to “decide everything first”

Neither is “right.” If you’re writing flight control software, spec-driven is overwhelmingly correct. If you’re exploring an unknown market in a startup, exploration-driven is overwhelmingly correct. The problem is that humans are not good at switching styles based on context.

It’s Not About Skill

The gap between these two styles is independent of skill level. A 10-year veteran stays spec-driven if they’re spec-driven; a new hire moves exploration-driven if they’re exploration-driven. With experience, some become able to do both, but the default tendency is remarkably stable.

This is a stable individual difference measurable by personality tests. Let’s look at the structure.

Another Important Axis — Need for Cognition

But there’s an important caveat. Work style has a second axis independent of Tolerance for Ambiguity: the motivational dimension of how much someone enjoys versus finds painful the act of thinking itself. In psychology, this is called Need for Cognition (NFC), a construct introduced by Cacioppo & Petty (1982), with nearly 80 years of research behind the TA tradition but treated as a separate lineage of individual differences⁷.

Add this second axis and you theoretically get a 2×2 with four types. In practice, however, one of those cells (low NFC + low NFCC, not covered in this article) is extremely rare in structured labor markets⁸, and in the context of labor theory, three types is the most natural classification.

Type	Thinking motivation (NFC)	Tolerance for Ambiguity	Traits
Spec-driven (clean-work)	High	Low (high NFCC)	Strong internal standards, wants to reach an answer
Exploration-driven	High	High (low NFCC)	Thinks while accepting ambiguity, holds multiple options
Instruction-waiting (thought-delegating)	Low	—	Avoids thinking itself, delegates judgment

The Tolerance for Ambiguity (TA) and Need for Cognitive Closure (NFCC) discussed in detail in Chapter 2 onward are the axes that distinguish the first two types. The analysis of the instruction-waiting type along the “thinking motivation (NFC)” axis is detailed in a separate article: How Instruction-Waiting Workers Can Survive the AI Era. This article should be understood as a detailed reference for the psychological background of the two types among thinking people — spec-driven and exploration-driven.

2. Construct 1: Tolerance for Ambiguity

Frenkel-Brunswik’s Discovery

The concept of “tolerance for ambiguity” was introduced by the Austrian-born psychologist Else Frenkel-Brunswik in her 1949 paper Intolerance of Ambiguity as an Emotional and Perceptual Personality Variable³. She originally studied ethnocentrism in children and, in the process, realized that a specific cognitive style underlies prejudice.

That style was “intolerance of ambiguity.” She described this trait as³:

• Denial of emotional ambivalence (the coexistence of love and hate for the same object)
• Inability to tolerate cognitive incongruity
• Inability to recognize positive and negative features coexisting in one person
• Quick generalizations based on surface features
• Black-and-white dichotomous views of life

In other words, the tendency of “not being able to accept gray zones.” Frenkel-Brunswik further developed this concept in The Authoritarian Personality (1950), co-authored with Adorno, Levinson, and Sanford, positioning it as the core cognitive style of the authoritarian personality⁹.

This is a significant claim. It suggests that the cognitive trait of “not tolerating ambiguity” can ripple all the way into political and social attitudes.

Budner, MSTAT, and Modern Measurement

Later, Budner (1962) developed a 16-item scale, making quantitative measurement possible. The widely-used current scale is McLain’s MSTAT-II (1993, revised 2009), which measures three dimensions:

1. Discomfort with novelty
2. Discomfort with complexity
3. Discomfort with insolubility

Low Tolerance for Ambiguity has been repeatedly shown to correlate weakly with IQ. It’s not “smart vs. dumb” but “how the brain handles uncertain information.”

The Inescapable Ambiguity of Requirements Engineering

In software development contexts, this becomes especially tricky. Requirements engineering research shows that natural-language requirements specifications are fundamentally unavoidable in ambiguity¹⁰. Ambiguity in requirements is defined as “having multiple interpretations despite the reader’s knowledge of the requirements engineering context,” and removing it completely to an implementable level is, in principle, difficult.

Engineers with low tolerance for ambiguity are noted as prone to “not being able to progress unless every question is answered to the smallest detail”¹¹. For them, the ambiguity of requirements specifications is experienced as a psychological threat. So they can’t start. They ask endlessly. They voice anxiety about the undecided parts.

Labeling this “neurotic” or “indecisive” misreads the structure. Their brain’s threshold for processing uncertain information is set lower than others’.

3. Construct 2: Need for Cognitive Closure

Kruglanski’s Theory

Where Tolerance for Ambiguity measures “discomfort with uncertain information,” Need for Cognitive Closure (NFCC), systematized by Arie Kruglanski in the 1990s, captures a more motivational dimension¹².

Kruglanski and Webster defined NFCC as¹:

“A desire for definite knowledge on some issue, representing a dimension of stable individual differences as well as a situationally evocable state.”

They developed a 42-item scale (now commonly used as a 15-item short form) measuring five dimensions:

1. Need for Order
2. Need for Predictability
3. Decisiveness
4. Avoidance of Ambiguity
5. Closed-mindedness

“Seizing” and “Freezing” — A Two-Stage Cognitive Mechanism

Kruglanski described NFCC’s operating mechanism as a two-stage “seizing and freezing” model¹².

Seizing: High-NFCC individuals feel an intense urgency to reach a judgment or conclusion. They try to reach closure as quickly as possible by any means. They rush information processing and jump on the first clues they get.

Freezing: Once they reach a conclusion, they try to permanize that judgment. Even if new information arrives, they don’t update. The psychological cost of revision is too high.

This “seize and freeze” cycle is the essence of Need for Cognitive Closure.

Specific Behavioral Effects

Multiple studies show characteristic behavior patterns of high-NFCC individuals:

• Strong risk aversion: Schumpe et al. (2017) showed that people higher in NFCC take fewer economic risks and discount delayed rewards more¹³
• Premature judgment freezing: They reach conclusions on limited information, with subsequent revision being difficult¹²
• Resistance to novelty: They prefer existing routines and avoid switching to new methods¹²
• Correlation with authoritarianism/conservatism: Jost et al.’s (2003) meta-analysis (88 samples, 12 countries) showed a significant positive correlation (r=.26) between NFCC and political conservatism¹⁴

NFCC is the flip side of “following the rules” and “maintaining existing frameworks.” This can be socially functional or dysfunctional depending on context.

flowchart TB
    Trigger["Uncertain situation<br>(vague spec, unknown problem)"]
    Trigger --> Discomfort["Psychological discomfort"]
    Discomfort --> Seize["Seizing<br>Rush to a conclusion"]
    Seize --> Decision["Some conclusion<br>(spec nailed down to details)"]
    Decision --> Freeze["Freezing<br>Reject new information"]
    Freeze --> Behavior["Spec-driven work style<br>Wants to proceed only by plan"]

    classDef stress stroke:#cf222e,stroke-width:2px
    classDef action stroke:#0969da,stroke-width:2px
    class Discomfort,Freeze stress
    class Seize,Decision,Behavior action

4. Construct 3: Overlap and Differences with Perfectionism

A question may arise. “Isn’t this just perfectionism?”

Certainly there’s significant overlap. Perfectionism research shows that maladaptive perfectionism is characterized by “excessive fear of failure,” “unreachable standards,” and “rigidity” — descriptions that sound similar to NFCC and low TA on the surface. The psychological mechanics of perfectionism are detailed in The Psychology of Perfectionism, which you can read for more depth.

But important differences exist:

Construct	Core motivation	What it measures
Maladaptive perfectionism	Fear of failure, self-evaluative strictness	“I don’t want to make mistakes”
Low Tolerance for Ambiguity	Discomfort with uncertain information	“Gray areas feel wrong”
High Need for Cognitive Closure	Urgency for a definite answer	“I can’t stand things being undecided”

Perfectionists are motivated by “I don’t want to fail.” Low-TA / high-NFCC people, on the other hand, are motivated centrally by discomfort with “the uncertain state itself,” not concern about the completeness of work.

In practice the two often compound. The “only-wants-clean-work” phenomenon is probably produced by the overlap of these three psychological traits.

5. Cultural Factor — Japan’s Extreme Uncertainty Avoidance

Hofstede’s 92 — An Extreme Value

So far we’ve discussed individual psychology. But there’s another factor we can’t ignore: culture.

In Geert Hofstede’s cultural dimensions, Japan’s Uncertainty Avoidance Index is 92, placing it at the top tier worldwide¹⁵. This index measures “the degree to which ambiguous or unknown situations feel threatening.” Greece (112) and Guatemala (101) score higher, but 92 is the highest in East Asia, and Japan is described as “one of the most uncertainty-avoiding cultures in the world.”

According to Hofstede’s analysis, Japan has rituals and procedures for every life transition “from cradle to grave”¹⁵. Weddings, funerals, school openings and closings, business etiquette — proper forms of behavior are codified in detail. This isn’t individual NFCC but a collective tendency to compress uncertainty at the social-system level.

Japan’s SI Industry Contract Model

This cultural tendency is strongly reflected in Japan’s software industry. Japanese system integrators (SI) still dominate with “fixing requirements before signing a fixed-price contract” waterfall models. This is structurally locked in by a triple overlap of:

• Customer uncertainty avoidance (wanting to transfer risk to the vendor)
• Vendor uncertainty avoidance (wanting to define the scope)
• Organizational psychology on both sides (managers can’t explain ambiguous states)

The phrase “dandori hachibu” (段取り八分, “80% of the work is preparation”) symbolizes the norm “preparation is the essence of work” — born from the same cultural soil.

So in Japan, feeling “I only want to do clean work” is both an individual trait and a culturally trained cognitive pattern. Compared to someone with the same trait in the West, this is seen as “normal” — even conforming to cultural norms — rather than abnormal. An important difference.

6. The AI-Era Paradox — Dramatic Cost Reduction for “Just Build It”

The Arrival of Vibe Coding

Now to AI.

In early 2025, Andrej Karpathy began using “vibe coding” on X (formerly Twitter). It describes a style where you don’t write code line by line but convey intent to an AI in natural language, test the generated code, and revise with follow-up prompts⁴.

In Karpathy’s own phrase, you develop “forgetting that the code even exists.” Y Combinator reported in March 2025 that 25% of its Winter batch startups had codebases that were 95% AI-generated⁴. “Vibe coding” was named Collins English Dictionary’s 2025 Word of the Year.

The fundamental change here: the cost of “just build and try” has dropped by orders of magnitude.

The Reversal of Exploration and Exploitation Costs

James March’s 1991 classic framework of “exploration and exploitation” in organizational learning clarifies the significance of this change⁵.

March classified organizational learning into two modes:

• Exploitation: Refining known methods, improving efficiency. “Refinement, choice, production, efficiency, selection, implementation, execution”
• Exploration: Trying new possibilities. “Experimentation, risk taking, variation, flexibility, discovery, innovation”

March’s central insight: these two modes have a tradeoff. Organizations have to allocate limited resources between them, and investing in exploration reduces short-term efficiency while over-focus on exploitation reduces long-term adaptability.

“Spec-driven” leans toward exploitation; “exploration-driven” leans toward exploration. Before AI, the unit cost of exploration was high. Prototypes took time; writing throwaway code was a luxury. So “nail down the spec before building” was a rational call.

But AI dramatically dropped the unit cost of exploration. You can build a prototype in 30 minutes and throw it away. You can try 10 implementations in parallel and compare them. You can feel out requirements with running code before finalizing them.

flowchart TB
    subgraph Before["Before AI"]
        E1["Exploration cost: High<br>(prototypes take time)"]
        X1["Exploitation cost: Medium"]
        E1 --> Logic1["Spec-driven is rational<br>Decide before building"]
        X1 --> Logic1
    end

    subgraph After["AI era"]
        E2["Exploration cost: Dramatically low<br>(prototypes in minutes)"]
        X2["Exploitation cost: Reduced"]
        E2 --> Logic2["Exploration-driven advantage<br>Build, test, fix"]
        X2 --> Logic2
    end

    Before --> After

    classDef oldStyle stroke:#8957e5,stroke-width:2px
    classDef newStyle stroke:#2ea44f,stroke-width:2px
    class Logic1 oldStyle
    class Logic2 newStyle

This isn’t about “spec-driven is obsolete.” It’s that the premises of spec-driven rationality have changed. The cost/benefit of pre-planning has structurally declined.

7. Do “Rule-Following People” Lose to AI?

Back to the main question. Are people who can’t move without full specs at a disadvantage in the AI-era labor market?

Intuitively, yes — because executing structured tasks is exactly what AI is best at. If specs are fully clear, AI can execute them; the argument is persuasive.

But the data complicates the picture.

The Brynjolfsson-Li-Raymond Paradox

Erik Brynjolfsson, Danielle Li, and Lindsey Raymond’s large-scale study (2023 NBER working paper, published in Quarterly Journal of Economics 2025) analyzed the introduction of a generative AI assistant to 5,172 customer support agents¹⁶. The results were unexpected:

• 15% overall improvement in issues resolved per hour
• 34% improvement for novices and low-skilled workers
• Almost no change for experienced / high-skilled workers

This finding, finalized in the QJE 2025 version, has become a key reference point in AI labor economics¹⁶.

What’s happening? The researchers interpret it as “AI transmits best practices from better workers, helping newer workers descend the experience curve faster.” AI acts as an accelerator for structured rule execution.

In other words, the “rule-following” style doesn’t disappear with AI — it may actually be reinforced. AI generates or organizes specs and procedures, and humans execute. This division of labor might actually be a comfortable environment for low-TA / high-NFCC people.

Dell’Acqua’s Jagged Frontier

But the story doesn’t end there. Fabrizio Dell’Acqua et al. (2023) conducted a large-scale experiment with 758 BCG consultants using GPT-4⁶.

Results were two-faced:

• Tasks inside AI’s “frontier”: 12% higher completion, 40% higher quality, 25% higher speed
• Tasks outside AI’s “frontier”: Consultants using AI were 19 percentage points less likely to produce the correct solution compared to those without AI

The researchers called this irregular boundary the jagged frontier⁶. AI shines in some tasks and fails badly in others — tasks that look similar. And the boundary is context-dependent and uncertain until you try.

Riding this jagged frontier requires the skill to critically evaluate AI output and know when to trust and when to doubt. Dell’Acqua et al. called people good at this “Centaurs” (dividing tasks between human and AI) and “Cyborgs” (interweaving tightly).

Low-TA / high-NFCC people hit a weakness here. Judging the jagged frontier is fundamentally an ambiguous act. You’re constantly facing “AI might be right, might be wrong, and the boundary is context-dependent and uncertain.”

Spec-driven people tend to stick to “verify every AI output before accepting,” failing to draw out AI’s strengths. Exploration-driven people naturally think “just have AI build it, try something else if it doesn’t work.”

So: in work within the rules, AI complements you; but the meta-level judgment of “when to trust AI” requires tolerance for ambiguity. At this layer, low-TA people are structurally disadvantaged.

8. Conclusion — Both Traits Have Winning Paths

As we’ve seen, the “only-wants-clean-work” phenomenon has this structure:

flowchart TB
    Core["Only wants clean work"]
    Core --> P1["Low Tolerance for Ambiguity<br>Discomfort with uncertain info"]
    Core --> P2["High Need for Cognitive Closure<br>Motivation for clear answers"]
    Core --> P3["Maladaptive perfectionism<br>Fear of failure"]
    Core --> P4["Cultural factor<br>Japan's UAI 92"]

    P1 --> AI["AI-era implications"]
    P2 --> AI
    P3 --> AI
    P4 --> AI

    AI --> Strong["Strengths<br>Regulated / safety-critical<br>Procedural competencies"]
    AI --> Weak["Weaknesses<br>Jagged frontier judgment<br>Exploratory development"]

    classDef core stroke:#cf222e,stroke-width:3px
    classDef strength stroke:#2ea44f,stroke-width:2px
    classDef weakness stroke:#d29922,stroke-width:2px
    class Core core
    class Strong strength
    class Weak weakness

March’s exploration/exploitation tradeoff⁵ doesn’t disappear with AI either. Organizations still need both, and individuals can choose where their traits fit. Additionally, TA and NFCC have situational dependence¹, so flexibility can be increased through environmental adjustment and training.

The answer to “do people who only want clean work lose to AI?” is neither simply yes nor no. More precisely, each trait has its own winning path and its own traps:

• Spec-driven (low TA, high NFCC): The moment you aim your strong internal standards at AI as a “selection tool” rather than a “writing tool,” your weakness flips into competitive advantage. Brynjolfsson et al.’s research¹⁶ shows that structured execution is actually reinforced by AI in some ways
• Exploration-driven (high TA, low NFCC): Your parallel-holding capacity is exponentially amplified in the vibe coding and jagged frontier environment⁶. But watch out for the unique traps of infinite exploration loops and taste loss

Both traits have a “path to surviving the AI era without abandoning who you are.” But the specific tactics differ fundamentally between traits. This article focuses on “why we can say this” from the research evidence — for actual tactics, see the trait-specific playbooks below.

📘 For spec-driven types: The Spec-Driven Engineer’s AI-Era Playbook — Four tactics for weaponizing strong internal standards.

📗 For exploration-driven types: The Explorer’s AI-Era Playbook — Four tactics for exponentially accelerating parallel exploration.

📕 For instruction-waiting types, or managers of them: How Instruction-Waiting Workers Can Survive the AI Era — The three-stage shift (short-term booster, deepening dependence, replacement) and escape routes.

If you’re unsure which type you fit, try the 30-second self-diagnosis at the start of each article.

Summary

“I can only work when the specs are nailed down” and “I only want to do clean work” are not about intelligence or laziness. They can be explained by Tolerance for Ambiguity and Need for Cognitive Closure — stable cognitive traits with nearly 80 years of psychological research behind them.

In the AI era, these traits have both “strengthened” and “weakened” sides. Spec-driven types have a winning path if they use their strong internal standards as the “judge” role. Exploration-driven types have a winning path if they amplify parallel-holding capacity with AI. What both need is to not deny their traits, and to redesign how they approach AI based on those traits.

The crude prescription “in the AI era, everyone has to become comfortable with ambiguity” is wrong. The correct prescription is: understand your cognitive traits, and know the winning paths and traps that come with them. I hope this article serves as the research foundation for that understanding.

Related Articles

Trait-Specific Playbooks / Recognition Guide (Practical)

• The Spec-Driven Engineer’s AI-Era Playbook — Four tactics for the spec-driven type
• The Explorer’s AI-Era Playbook — Four tactics for the exploration-driven type
• How Instruction-Waiting Workers Can Survive the AI Era — The three-stage shift and escape routes for the instruction-waiting type

Related Psychology / AI-Use Articles

• The Psychology of Perfectionism: Between High Standards and Self-Destruction — Two faces of adaptive/maladaptive perfectionism
• The Opportunity Cost of “Only Using AI for What You Can Fully Review” — How perfectionism blocks AI adoption
• Why Process-Attached People Struggle With AI — Process vs. outcome motivation structures

References

1. Motivated Closing of the Mind: “Seizing” and “Freezing” - Kruglanski, A. W., & Webster, D. M. (1996). Psychological Review, 103(2), 263-283. 【Reliability: High】 ↩︎ ↩︎² ↩︎³ ↩︎⁴

2. Individual Differences in Need for Cognitive Closure - Webster, D. M., & Kruglanski, A. W. (1994). Journal of Personality and Social Psychology, 67(6), 1049-1062. 【Reliability: High】 ↩︎ ↩︎²

3. Intolerance of Ambiguity as an Emotional and Perceptual Personality Variable - Frenkel-Brunswik, E. (1949). Journal of Personality, 18, 108-143. 【Reliability: High】 ↩︎ ↩︎² ↩︎³

4. Vibe coding - Wikipedia - Source for the origin of the term (Karpathy 2025), Y Combinator statistics, and Collins Word of the Year 2025. 【Reliability: Medium】 ↩︎ ↩︎² ↩︎³

5. Exploration and Exploitation in Organizational Learning - March, J. G. (1991). Organization Science, 2(1), 71-87. 【Reliability: High】 ↩︎ ↩︎² ↩︎³

6. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality - Dell’Acqua, F., McFowland III, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2023). Harvard Business School Working Paper, 24-013. 【Reliability: High】 ↩︎ ↩︎² ↩︎³ ↩︎⁴

7. The Need for Cognition - Cacioppo, J. T., & Petty, R. E. (1982). Journal of Personality and Social Psychology, 42(1), 116-131. Initial paper on Need for Cognition (NFC). Scale measuring individual differences in the intrinsic tendency to enjoy effortful thinking. A foundational text cited over 8,600 times. 【Reliability: High】 ↩︎

8. Theoretically, a 2×2 classification yields four types, but the low-NFC (dislikes thinking) + low-NFCC (comfortable with ambiguity) combination is rarely observed in structured labor markets. People with this type tend to be pressured into instruction-waiting by the environment or distribute into unstructured environments (freelance, non-professional contexts), so in labor theory they’re effectively close to an empty set. Hence this article and the related series use three types. ↩︎

9. Ambiguity tolerance–intolerance - Wikipedia - Adorno, T. W., Frenkel-Brunswik, E., Levinson, D. J., & Sanford, R. N. (1950). The Authoritarian Personality. Harper. 【Reliability: Medium】 ↩︎

10. Ambiguity in Requirements Specification - Berry, D. M., & Kamsties, E. (2004). In Perspectives on Software Requirements. Foundational paper on the inescapability of ambiguity in requirements specifications. 【Reliability: Medium-High】 ↩︎

11. Behavioral Interview Questions for Assessing Dealing with Ambiguity in Engineering Roles - Practical framework for handling ambiguity in engineering roles. 【Reliability: Medium】 ↩︎

12. Cognitive and Social Consequences of the Need for Cognitive Closure - Kruglanski, A. W., & Webster, D. M. (1996). European Review of Social Psychology, 8, 133-173. 【Reliability: High】 ↩︎ ↩︎² ↩︎³

13. Need for Cognitive Closure decreases risk taking and motivates discounting of delayed rewards - Schumpe, B. M., Brizi, A., Giacomantonio, M., Panno, A., Kopetz, C., Kosta, M., & Mannetti, L. (2017). Personality and Individual Differences, 107, 66-71. Shows that higher NFCC is associated with greater risk aversion and steeper discounting of delayed rewards. 【Reliability: High】 ↩︎

14. Political Conservatism as Motivated Social Cognition - Jost, J. T., Glaser, J., Kruglanski, A. W., & Sulloway, F. J. (2003). Psychological Bulletin, 129(3), 339-375. Classic meta-analysis linking Need for Cognitive Closure to political conservatism (88 samples, 12 countries, r=.26). 【Reliability: High】 ↩︎

15. Uncertainty Avoidance and the Japanese - Penn State University global research article. Explains the link between Hofstede’s score of 92 and Japanese culture. Original data from Hofstede Insights. 【Reliability: Medium-High】 ↩︎ ↩︎²

16. Generative AI at Work - Brynjolfsson, E., Li, D., & Raymond, L. (2025). Quarterly Journal of Economics, 140(2), 889-942. Study of 5,172 customer support agents. 15% overall, 34% improvement for low-skilled workers. Originally published as an NBER working paper in 2023. 【Reliability: High】 ↩︎ ↩︎² ↩︎³