AI Is Undermining Leaders’ Judgment. Here’s What to Do About It.

Leonid Sudakov, Nathan Furr · article · 2026-08-19 source ↗ Reader 🖥️

practical · mixed — Opens theoretical/diagnostic (a named 'Original Judgment Paradox' backed by a field experiment and a three-stage erosion mechanism), then turns practical: two named methods (CAST, RED) with historical and client narratives as illustration and evidence of feasibility.

Structure

Key terms

• original judgment
  The uniquely human capacity to see beyond the dominant narrative and make choices that represent one's values under uncertainty — the thing that underlies competitive breakthroughs.
• breadth of perception
  One of two interlocking capacities behind original judgment: noticing weak signals and adjacent patterns beyond the obvious.
• independence of interpretation
  The other capacity behind original judgment: forming one's own view rather than deferring to the model's or the group's.
• cognitive offloading Cognitive Surrender
  The first stage of judgment erosion: leaders increasingly relying on data-driven inputs, visible even at the neural level.
• deference to tools
  The second erosion stage: trusting outputs from an 'algorithmic authority' the person does not fully understand.
• organizational monoculture
  The third erosion stage: an organization-wide convergence on the same metrics, models, and narratives because everyone is deferring to the same tools.
• structured curiosity (the CAST method)
  A disciplined practice — Collect from the edges, Arrange disparate signals, Suspend premature closure, Trigger alternative interpretations — for maintaining breadth of perception deliberately rather than as a personality trait.
• intentional dissent (the RED method)
  A disciplined practice — Recruit independent perspectives, Engage through structured challenge, Dialogue to sustain over time — for cultivating dissent as an organizational capability rather than hoping for spontaneous bravery.

Propositions

• explicitcontestedAs organizations invest more in AI-driven decision-support infrastructure, their leaders are being trained out of original judgment — the capacity that creates competitive advantage — even though it underlies nearly every story of competitive breakthrough.conf: high
  extendsDesign AI Systems That Actually Strengthen Human Reasoning · p11 — Explicitly cited in the text: this article picks up where the design-literature piece leaves off, taking up 'a prior question' — how leaders cultivate judgment no interface redesign can supply.
  extendsCognitive Surrender · p1 — Original-judgment erosion is a leadership-specific, three-stage elaboration of the cognitive-offloading-to-cognitive-surrender continuum: offloading gives way to full deference once organizational monoculture forms.
  contradicts ←This Week in AI: When agents outnumber people · p5 — This piece asserts foundational knowledge becomes more valuable and (implicitly) will be maintained as a check on AI output, while the other argues that heavier reliance on AI decision-support is actively training leaders' original judgment out of them — the same trend (more AI in decisions) is read as building versus eroding the human capacity needed.
• explicitIn a field experiment with 228 experienced evaluators assessing 48 submissions to an MIT innovation challenge, when the AI tool recommended rejecting submissions the expert panel had approved, evaluators largely followed the AI's rejections, filtering out high-potential innovations.conf: high
• explicitcontestedEvaluators given a narrative explanation for the AI's recommendation were less likely to override an incorrect AI recommendation than evaluators given the recommendation with no explanation at all — the explanation made judgment worse, not better.conf: high
  contradictsWhy Great Leaders Need More Than Good Judgment · p9 — Goldsmith recommends deliberately creating AI personas, including contrarian ones, to surface a leader's blind spots. This article's central finding — that AI-supplied recommendations measurably suppress independent override judgment, and that explanation makes this worse — suggests AI personas risk reproducing the very algorithmic-authority deference this article warns against rather than curing it.
  supportsThe Psychological Costs of Adopting AI · p3 — This article's field-experiment finding is a specific, empirical instance of the general cognitive-offloading-undermines-independent-understanding mechanism Storey's piece describes.
• explicitOriginal judgment erodes through three sequential stages: cognitive offloading (relying on data-driven inputs), deference to tools (trusting algorithmic authority without understanding its logic), and organizational monoculture (a shared set of metrics, models, and narratives across the organization).conf: high
  supportsDesign AI Systems That Actually Strengthen Human Reasoning · p9 — 'Organizational monoculture' is the leadership-level, systemic manifestation of the idea-homogenization mechanism Reganti documents at the individual AI-output level (LLMs fixate on the first idea and average toward the mean).
  supportsPresentation: The Ironies of A^2 I^2 (Ironies of Automation and Artificial Intelligence in Incidents) · p3 — Same Bainbridge-style irony — automation erodes the practice that builds the judgment it's meant to augment — reapplied from incident response to strategic/creative leadership judgment.
• explicitForecasting accuracy depends less on the volume of data collected than on the diversity/variety of its sources.conf: medium
• explicitDissent rarely survives long enough to be tested without deliberate social reinforcement — most dissenting ideas fail not because they are wrong but because their originators become isolated before evidence can accumulate.conf: high
• explicitThe problem is not that leaders use AI and data — they should — but that sophisticated decision-support systems are training leaders to defer rather than originate.conf: high
• implicitAI-generated recommendations carry a more insidious form of authority than ordinary hierarchical deference, because their veneer of objectivity makes disagreement look irrational in a way a fallible human boss's stated opinion does not automatically produce.conf: medium
  extends12) Greger Teigre Wedel: Når sjefen har svaret, tier resten · p1 — Wedel's 'when the boss already has the answer, other voices go quiet' mechanism, reapplied to AI: an algorithm's veneer of objectivity produces the same chilling effect on dissent that a visibly answer-holding human boss produces.
• implicitIndividual and small-group practices (CAST, RED) can meaningfully counteract an erosion process the article itself frames as systemic and infrastructural (decision-support investment, organization-wide monoculture) — the prescribed remedy operates at a different level of scale than the diagnosed cause.conf: medium

Arguments

Narratives

• MIT/Harvard field experiment: 228 evaluators judging 48 innovation-challenge submissions under three conditions (human-only, AI-with-narrative-explanation, AI-with-no-explanation), compared against expert judges.
  → Supports the core erosion thesis and specifically the counterintuitive finding that explanation worsens rather than improves override behavior.
• Sam Walton building a discount retailer in towns under 50,000 people against conventional wisdom; Steve Jobs integrating calligraphy into the personal computer; Jensen Huang betting on unproven hardware emulation at Nvidia.
  → Supports the claim that original judgment underlies competitive breakthroughs.
• Tu Youyou, without a PhD, returning to 2,000 ancient Chinese medical texts after Western pharmaceutical screening pipelines were exhausted, and noticing a fourth-century manuscript detail (extract cold, not boiling) that led to artemisinin.
  → Supports 'structured curiosity' as a disciplined, teachable practice rather than an innate trait — grounds the CAST method.
• Pathologist Robin Warren and trainee doctor Barry Marshall building a deliberate dissent network (100-patient case series, outside methodological critics, skeptical-venue publication) to overturn the stress-and-acid consensus on peptic ulcers and establish H. pylori as the cause.
  → Supports 'intentional dissent' as requiring deliberate social reinforcement to survive — grounds the RED method.
• An anonymized global company transforming from scale manufacturing to a health-focused ecosystem ran a nine-month program: 15 high-potential leaders in five mismatched teams of three, given complex out-of-expertise challenges with no predetermined solution.
  → Supports that structured curiosity can be organizationally engineered, not just individually practiced.
• Two anonymized global companies built advisory networks of external experts outside reporting lines, tasked with pressure-testing logic rather than providing answers, which shifted over time from polite questioning to direct critique.
  → Supports that intentional dissent can be organizationally engineered as a protected practice space.

Questions addressed

• ✅ How can leaders preserve or cultivate original judgment in an AI-saturated decision environment?
  Two deliberate practices: structured curiosity (CAST) and intentional dissent (RED).
• ⭕ Why does explaining an AI recommendation (rather than presenting it bare) make people less likely to catch its errors?

Gaps

• incompleteThe article reports as 'striking' that narrative AI explanations reduced override of incorrect recommendations, but never investigates or theorizes the mechanism — it doesn't engage with why explanation would increase rather than decrease inappropriate reliance, which runs counter to the common assumption (implicit in much explainable-AI work) that explanation supports appropriate calibration of trust.
  Read next: Explainable-AI and automation-bias research specifically on when and why explanation increases versus decreases over-reliance on an AI recommendation.
• incompleteCAST and RED are each illustrated by exactly one historical pair plus one or two anonymized client engagements, with no measurement of whether these practices actually preserved judgment quality, no discussion of failure cases, and no account of whether the featured companies were simultaneously reducing AI usage (making it unclear whether the intervention, rather than reduced AI exposure, produced the reported outcomes).
  Read next: Any controlled or longitudinal study of structured-curiosity/structured-dissent interventions specifically in AI-saturated decision environments.
• uninformedThe piece never engages the tension it stands in with 'broaden your inputs' leadership advice — e.g. work recommending leaders deliberately create AI personas, including contrarian ones, to surface blind spots. If AI-supplied recommendations measurably suppress independent override judgment (this article's own central finding), then leaning on AI personas for a 'different view' risks reproducing the very algorithmic-authority deference this article warns against, rather than curing it.
  Read next: extract-why-great-leaders-need-more-than-good-judgment (Marshall Goldsmith's 'panoramic leadership,' which recommends AI personas as a blind-spot tool)
• incompleteThe article frames judgment erosion purely at the level of the individual leader (a leader deferring to AI). It does not address the cascading, accountability-laden version of the same problem: subordinates who must explain and defend a leader's AI-endorsed decision to others, and who may mask, embrace, or reinterpret that decision to protect their own credibility — a distinct organizational dynamic from the leader's own judgment eroding.
  Read next: extract-when-employees-are-held-accountable-for-ai-generated-decisions

Closes gaps elsewhere

• Why Mentoring Matters More in the AI Era · g4 — Corroborates, via a different causal path, that the gap's underlying assumption is false: this article's field experiment and three-stage erosion mechanism show today's leaders' own judgment actively degrading right now, through AI-induced cognitive offloading and deference — not only through the mentoring piece's concern about tomorrow's leaders lacking early-career repetition. Either way, the 'stable reservoir of tacit judgment' the mentoring piece assumes is not stable.
• The founder of Postman says you have to kill your existing org chart · g3 — Defines the undefined term directly: judgment is specifically breadth of perception plus independence of interpretation, developed through two named, teachable practices (structured curiosity / CAST and intentional dissent / RED) rather than left as an unelaborated slogan.

Related documents

• 12) Greger Teigre Wedel: Når sjefen har svaret, tier resten
• Presentation: The Ironies of A^2 I^2 (Ironies of Automation and Artificial Intelligence in Incidents)
• Cognitive Surrender
• Design AI Systems That Actually Strengthen Human Reasoning
• Building an AI Incident Copilot: How I Automated the First 15 Minutes of Every Production Incident
  Capacity to Verify AI Output Ironies of Automation
• Det skal være vanskelig
  Cognitive Surrender Ironies of Automation
• Why "good enough" is the whole point of 80/20
  Cognitive Surrender Decision-Making Practices
• 16) Inga Strümke om hvor AI er på vei, den outsourcede problemløsningen og datamanifolden
  Cognitive Surrender Explainable AI
• The Psychological Costs of Adopting AI
  Cognitive Surrender Explainable AI
• Research: Why Some Junior Employees Work Well with AI—and Others Don’t
  Capacity to Verify AI Output Cognitive Surrender