1. Gemini
**The knowledge your model will never read**
Organizations aiming to leverage Artificial Intelligence often overlook a crucial element: the knowledge that resides within their own walls, much of which remains uncodified and deeply personal. According to an op-ed by Robert Garita published in on August 10, 2026, AI's true impact is not in transforming organizations but in amplifying what they already are. The challenge, Garita suggests, lies in firms not knowing where their vital organizational knowledge truly lives, particularly the 'tacit' kind that AI models struggle to access.
Drawing on Michael Polanyi's 1966 observation that "we know more than we can tell," the op-ed differentiates between two types of knowledge. Explicit knowledge is readily coded and found in manuals, policies, databases, and contracts. Tacit knowledge, however, is embedded in people – their judgment, accumulated experience, craft intuition, relationships, and ability to read context. This personal, often unarticulated knowledge is frequently what dictates critical decisions, such as an operations chief discerning a hidden problem behind a seemingly positive KPI, or a negotiator instinctively knowing when to conclude discussions.
The scale of this tacit knowledge is significant, yet IDC reports there is no reliable measure of its share, a lack of clarity that itself constitutes a strategic risk. Externalizing this deep-seated knowledge is far from a simple task, certainly not a "two-day workshop." The article cites Nonaka and Takeuchi's 'Knowledge-Creating Company' and the Matsushita Corporation's experience in the 1980s. The company struggled to mechanize bread kneading until software engineer Ikuko Tanaka spent a year apprenticed to an Osaka International Hotel master baker, subsequently translating a specific stretch movement into an engineering specification.
Ignoring this vital human dimension can lead to substantial financial repercussions. The op-ed highlights Zillow as a stark example; the company possessed excellent explicit housing data and developed an algorithm to buy and flip homes. However, it lacked the critical local agent judgment, leading to the closure of the unit in November 2021 after experiencing a loss of $422 million in a single quarter and resulting in approximately 25% staff reductions.
This challenge is not isolated to single companies. A 2025 review by MIT Project NANDA of over 300 enterprise generative AI initiatives revealed that 95% yielded no return, with barriers primarily organizational rather than related to model quality. Furthermore, IDC indicates that 89% of organizations report some form of data-quality problem, contributing to more than half of all AI initiatives stalling after their pilot phase.
To truly harness AI's potential, organizations must first systematically identify critical knowledge, determine who possesses it, and understand how it is used and transferred. This structured approach, Garita argues, is fundamentally a matter of strategy, culture, process, and governance, rather than merely an IT ticket. While Costa Rica boasts talent and a national strategy, firms within the country, like those globally, still need to meticulously ascertain what knowledge they actually possess to succeed in an AI-driven future.
2. Grok
1) What matches the source The GEMINI ARTICLE follows the source excerpt almost point-for-point: Robert Garita op-ed, 10 August 2026, the “AI amplifies what organizations already are” framing, Polanyi quote, explicit vs. tacit distinction with the same two decision-making examples, Matsushita bread-kneading case (Ikuko Tanaka, Osaka hotel baker, one stretch movement), Zillow $422 million quarterly loss and ~25 % staff cut in November 2021, MIT Project NANDA 2025 review (300+ initiatives, 95 % no return, organizational barriers), IDC 89 % data-quality problems and >50 % of initiatives stalling after pilot, the “not a two-day workshop” line, and the final emphasis on strategy/culture/process/governance rather than an IT ticket plus the Costa Rica reference.
2) What is simplified or uncertain The article smooths the source’s phrasing into more fluid English without adding new facts. The only minor uncertainty is the repeated attribution to “IDC” for the tacit-knowledge measurement gap; the source mentions it once in passing and does not elaborate on methodology or year. All numerical claims are taken verbatim from the excerpt.
3) What is missing No independent verification or primary sources for the MIT NANDA or IDC statistics. No discussion of the actual cost, time, or measurable ROI of systematic tacit-knowledge capture. No examples of organizations that succeeded at this step. No details on Costa Rica’s national AI strategy or how firms there are (or are not) addressing the issue. No consideration of hybrid approaches that combine partial externalization with AI augmentation.
4) Your own opinion The core argument is sound: organizations routinely overestimate how much of their critical know-how is already explicit. The Zillow case is a useful, concrete illustration. At the same time, the piece leans heavily on dramatic failure statistics without sourcing or counter-examples, which weakens its prescriptive force. The reminder that knowledge strategy precedes technology deployment is valuable, but it stops short of offering any operational next steps.
3. Claude
Verification Analysis
1. Factual Accuracy Against Source
The Gemini summary is **highly faithful** to the source material. Every major claim traces back:
- Robert Garita op-ed,, August 10, 2026 ✓
- Polanyi 1966 quote "we know more than we can tell" ✓
- Explicit vs. tacit knowledge taxonomy ✓
- Operations chief/negotiator examples ✓
- Matsushata bread-kneading case: Ikuko Tanaka, year-long apprenticeship, Osaka International Hotel baker, one stretch movement ✓
- Zillow: $422M single-quarter loss, Nov 2021 closure, ~25% workforce reduction ✓
- MIT Project NANDA 2025: 300+ initiatives, 95% no ROI, organizational barriers ✓
- IDC: 89% data-quality issues, >50% stall at pilot ✓
- "Not a two-day workshop" phrase ✓
- Costa Rica talent/national strategy reference ✓
No fabricated statistics or misattributed claims detected.
2. Presentation Choices
The summary opts for narrative flow over the source's terse bullet style but introduces **no new interpretation**. One subtle shift: the source says AI "amplifies what organizations already are" as context for why tacit knowledge matters; the summary makes this the thesis statement. The framing is stronger but remains faithful.
The IDC "no reliable measure" point is mentioned once in the source; the summary doesn't overstate it but also doesn't flag that IDC's specific reports aren't cited.
3. Critical Gaps
**No counterpoint or nuance**: Are *all* tacit knowledge types equally critical? The Matsushita example succeeded—how common is that pathway?
**No primary source trail**: MIT "Project NANDA" is unfindable in 2025 literature (future date notwithstanding). IDC statistics lack report titles or dates.
**No cost-benefit analysis**: If externalizing tacit knowledge is expensive and slow, when does it beat alternative strategies like retaining key personnel or building AI co-pilots that learn *with* experts rather than replacing them?
**Costa Rica mention is a non-sequitur** without explaining *how* the national strategy addresses tacit knowledge specifically.
4. Verdict
**Reliable summary of the source; unverified source claims.**
The Gemini text accurately reflects what Garita wrote. The underlying op-ed, however, relies on compelling but unsourced statistics (particularly the MIT study) and a single cautionary tale (Zillow). The argument—that organizations underinvest in knowledge externalization before deploying AI—is **conceptually sound and well-illustrated**, but the piece reads more as a provocative warning than a rigorous evidence base.
For decision-makers: treat this as a useful framework for asking "what do our people know that our systems don't?" rather than as definitive proof that 95% of AI projects fail for this specific reason.
4. ChatGPT
1) **My Own Take on this Story**: The argument presented in the piece emphasizes the often-overlooked value of tacit knowledge within organizations, a crucial element for effective AI integration. While it's true that many companies possess vast resources of explicit knowledge, the unarticulated, experiential insights of employees can be far more impactful in decision-making and operational strategies. The Zillow example highlights the disastrous consequences of disregarding this human expertise; the failure stems not from a lack of data but from an inability to interpret and apply that data without the nuanced judgment that comes from experience. This narrative serves as a reminder that before investing in advanced technologies like AI, organizations must first develop strategies focused on identifying, codifying, and leveraging the priceless, experiential know-how held by their workforce.
2) **Where the Three Voices Agree**: - All three voices recognize that the Gemini summary is largely faithful to the source material, accurately representing key concepts, statistics, and examples provided by Garita. - They all point out the absence of sourced statistics regarding MIT's Project NANDA and the IDC claims, highlighting a critical gap in the evidence supporting the argument. - The emphasis on the importance of tacit knowledge versus explicit knowledge is a common thread, with all voices reinforcing that organizations often overlook this distinction, leading to strategic risks and developmental challenges.
3) **Where They Differ**: - **Tone and Interpretation**: While GEMINI emphasizes the practicality of the knowledge externalization process, GROK appears slightly more critical, mentioning the piece's reliance on failure statistics without providing success examples. In contrast, CLAUDE presents a more balanced view, acknowledging the summary's accuracy while highlighting that it lacks nuance and counterpoints. - **Operational Next Steps**: GEMINI does not offer any operational next steps or solutions, merely stating knowledge strategy is essential. In contrast, GROK critiques this absence, arguing the summary should provide more actionable insights.
4) **What to Check in the Original**: - Verification of the MIT Project NANDA study details, including specific titles, methodologies, and publication dates to assess the accuracy of its claims. - Examination of how the IDC statistics were derived, particularly concerning their assertions about the impacts of data quality issues on AI initiatives. - Further details on Costa Rica's national AI strategy to understand how the country is addressing the issues of tacit knowledge and whether any specific initiatives or frameworks have been established. - Additional context around the Zillow case—are there lessons to be learned about integrating tacit knowledge into decision-making processes, and how did other organizations potentially succeed in this area?