How your AI strategy might be making employees stop thinking

Generative AI use in workplaces risks creating widespread “knowledge decay”

We’re seeing a strange paradox right now. Generative AI can automate work at incredible speed, but most of what it produces isn’t good. Matthias Holweg from Oxford’s Saïd Business School and analyst Thomas H. Davenport describe the growing dependence on this low-quality AI content as “knowledge decay.” It’s when teams stop thinking critically because they trust the machine too much. The output looks polished, but underneath, judgment and originality begin to fade.

For executives, this should be a red flag. When employees rely on AI templates instead of their own reasoning, they lose the edge that keeps your business innovative. Over time, quality assurance weakens and trust in internal systems erodes. Processes designed for human input start breaking down because they’re filled with content no one truly understands or owns.

That’s where leadership makes the difference. AI is powerful, but it must remain a partner. The goal should be augmentation, using AI to speed up factual work. When people let AI handle too much, they end up accepting mediocrity disguised as efficiency. Maintaining high standards means ensuring employees still question, verify, and improve what AI delivers.

Verification, validation, and entropy are the central challenges in preventing organizational knowledge decay

Holweg and Davenport identify three challenges business leaders can’t ignore: verification, validation, and entropy. Each one strikes at the core of how organizations create and protect knowledge.

Verification means being able to tell what’s real and what’s automated. It’s more difficult than it sounds. For instance, companies are already facing hiring issues where candidates use AI to generate CVs or even real-time interview responses. The resumes look sharp, but may not reflect actual skills. This forces recruiters to spend more time validating authenticity through in-person conversations, time that automation was supposed to save.

Validation is about confirming where human intelligence truly adds value. Imagine a consulting team using AI for client reports. If those insights are entirely AI-generated, the client isn’t paying for thought leadership, they’re paying for automation. Teams will need to demonstrate what parts of the deliverable came from human insight and which were AI-assisted. That separation of value is critical for credibility.

Entropy is the slow drift of truth. AI systems generate text by predicting likely words. Each time an AI-generated document feeds into another AI model, the content moves further from the accurate source data. Over many cycles, knowledge becomes unreliable. In complex businesses, this could distort forecasts, compliance reports, or market analyses without anyone noticing until it’s too late.

The nuance here is balance. These risks don’t mean AI should be minimized, they mean it needs governance. Leaders should design workflows where humans review and challenge AI outputs. Real efficiency comes from structured human oversight that catches drift early and ensures data integrity remains intact.

In short: if AI is going to make your organization faster, make sure it’s also making it smarter.

Generative models risk “inbreeding” when trained on synthetic data, potentially undermining accuracy and diversity

A growing problem in AI development is “model collapse.” Matthias Holweg and Thomas H. Davenport describe this as what happens when generative models train on their own outputs or other synthetic data. The model begins reinforcing its previous results instead of learning from fresh, factual information. Over time, diversity in responses declines, and accuracy deteriorates. It stops learning from the real world and starts cycling through its own distortions.

This is harmful for businesses that depend on accurate data to make decisions. AI tools built on recycled content become unreliable when generating reports, recommendations, or summaries. Corporate strategies based on flawed models risk steering operations in the wrong direction. For executives, this is an operational one, potentially affecting forecasting, market intelligence, and compliance accuracy.

Leaders should ensure their teams understand that AI quality depends on the quality of its training data. Continuous retraining with verifiable, human-curated datasets keeps outputs relevant and reduces drift. Treat synthetic data as a temporary tool. For long-term competitiveness, AI systems must stay connected to the truth of real customer inputs, employee insights, and verified industry information. Maintaining that connection is a leadership responsibility that will define which organizations stay ahead as AI systems evolve.

Enterprises must redefine the use of AI to ensure it supplements rather than supplants human expertise

Holweg and Davenport argue for clear boundaries in how organizations apply AI. They advise using AI only when it demonstrably enhances outcomes. That means executives need to identify where automation provides measurable value and where human skill delivers a unique advantage. For example, automating the formatting of a report saves time, but generating the strategic conclusions within that report should remain human-driven.

This shift in thinking is essential. When AI use is unrestricted, employees may delegate too much, producing results that are fast but shallow. Establishing clear use rules protects intellectual integrity and keeps employees accountable for what they produce. In hiring, structured forms that ask for concrete details, roles, project results, budgets managed, reduce the risk of AI-generated nonsense. In client relations, verifying when and how AI was used preserves trust and transparency.

For business leaders, the nuance lies in governance. Restricting AI use isn’t about limiting innovation, it’s about making sure innovation adds real value. Transparency around AI involvement also matters. When employees document how AI assisted their work, it helps leadership understand where efficiencies are gained and where human expertise still needs to take the lead.

In practice, this means redefining AI policies to integrate critical oversight, ethical guidelines, and training. Managers should be prepared to challenge over-automation and recognize when human insight should take control. True progress will come from precision in how AI is applied, not from using AI everywhere without purpose.

Enterprise-wide governance and meticulous data tracking are essential for preserving “ground truth” integrity

Holweg and Davenport emphasize that organizations must establish clear systems to protect their factual data, what they call the “ground truth.” When AI is used to summarize or enhance written material, leaders need to ensure that every automated output connects back to verifiable, human-sourced information. Without strong data lineage, the organization risks losing clarity about what’s factual and what’s synthetic.

For executives, the practical implication is straightforward: data control is now a strategic function. When customer interviews, research insights, or financial inputs are processed by AI systems, companies must record not only the result but also the original material it was derived from. This allows for auditing, verification, and learning from the real evidence behind decisions.

Strong governance frameworks can transform AI from a compliance risk into a trust asset. Executives should define ownership over data quality, ensure employees understand documentation standards, and require traceability for all AI-modified content. This builds an internal system of accountability where every decision can be tracked back to authentic, provable data.

The nuance here is timing and consistency. Governance must evolve as fast as the technology it oversees. Executives who treat data management as a static process will fall behind. Maintaining “ground truth” requires continuous oversight, revision of standards, and a strong commitment to transparency. Enterprises that master this will not only protect accuracy but also strengthen trust across teams, regulators, and customers.

Businesses should adopt smaller, proprietary AI models rather than relying on large, generic public LLMs

Holweg and Davenport point out that public large language models (LLMs) often produce generic or inaccurate results, which limits their usefulness in specialized business environments. Smaller, proprietary models, customized using internal company data, offer better alignment with organizational needs. They can capture company-specific context, terminology, and processes that large models typically miss.

For executives, the key takeaway is control. Large public models may deliver convenience, but they are trained on general data that doesn’t reflect your company’s unique insights or tone. Smaller models trained on curated internal data are more reliable, more secure, and better suited for automating repetitive tasks without introducing external biases or errors. They can also reduce the risk of exposing sensitive information through open systems.

Adopting proprietary AI models isn’t just a technical choice; it’s a strategic one. It allows businesses to keep their competitive knowledge in-house while improving efficiency. Companies can decide what data trains their systems, ensuring compliance and accuracy. This approach helps align AI outputs with brand standards, regulatory expectations, and customer needs.

The nuance to recognize here is scalability. Smaller, domain-specific models don’t need to compete with the size of public LLMs, they need to excel in relevance and precision. Executives should support investments in building or fine-tuning custom AI that captures the voice and logic of their enterprise while maintaining rigorous performance benchmarks. In doing so, they move from simply using AI technology to owning it in a way that safeguards both capability and trust.

Failure to manage AI proliferation may repeat the historical “productivity paradox”

Holweg and Davenport warn that organizations that deploy generative AI without discipline could face the same stagnation businesses experienced in the early corporate computing era. Back then, major investments in automation didn’t produce expected productivity gains because integration was uncontrolled and poorly managed. The same risk exists now. Businesses that integrate AI tools without governance could see speed increase but quality decline, resulting in inefficiency masked by automation.

Executives should treat uncontrolled AI adoption as a systemic issue, not a technical one. Speed alone isn’t productivity if the underlying output loses accuracy, authenticity, or trust. When AI begins to flood internal processes, marketing, data analysis, recruitment, reporting, without careful oversight, quality assurance becomes reactive rather than proactive. This undermines long-term competitiveness and credibility.

The core issue is leadership attention. Governance frameworks must define where and how AI adds measurable value to operations. That means balancing automation benefits against the costs of monitoring accuracy and retraining employees to adapt to new tools. Without strategic oversight, AI proliferation can strain workflows, confuse accountability, and dilute decision quality.

The nuance here is alignment. Executives should focus on aligning AI deployment with the company’s operational objectives, productivity metrics, and human capital strategies. Intelligent AI use should enhance human decision-making, not obscure it. Firms that manage AI strategically will achieve real productivity gains, those that do not may see diminishing returns hidden behind a façade of digital progress.

Integrating human and AI capabilities, blending “human capital” with “token capital”

Microsoft CEO Satya Nadella describes the future of AI-enabled organizations through the balanced use of “human capital” and “token capital.” Human capital represents judgment, creativity, pattern recognition, and relationships. Token capital represents the machine learning systems and AI tools that augment human ability. When both are strategically connected, they create a learning loop, humans guide AI to new levels of effectiveness, and AI, in turn, amplifies human intelligence by analyzing outcomes at scale.

Executives should view this integration as the foundation of sustainable progress. Human insight defines the problems; AI improves the solutions. Nadella notes that every workflow enhanced by AI generates a better training signal, storing institutional knowledge more efficiently and allowing teams to query insights faster. Measured correctly, this relationship reduces redundancy and improves operational intelligence across the enterprise.

For leadership, the objective isn’t just deploying AI, it’s designing a continuous learning process. Establish internal benchmarks to test how well AI improves defined outcomes, such as speed, accuracy, and cost reduction. The most successful organizations will keep clear metrics to measure improvement while ensuring human judgment remains central to interpretation and oversight.

The nuance here is discipline. AI alone doesn’t create strategic advantage, how it’s directed does. C-suite leaders should make sure internal feedback systems allow human expertise to refine AI behavior continuously. This creates institutional memory that is both high in value and cost efficient. Companies that execute this approach correctly will accelerate innovation, maintain control over intellectual growth, and safeguard the distinctive knowledge that defines their competitive edge.

Recap

AI is now a permanent part of how businesses operate. The advantage will go to leaders who treat it as a disciplined partnership. Generative systems can move fast, but they can’t think critically, understand context, or recognize nuance, the things that drive real progress. Those strengths still belong to people.

Every organization faces a choice: use AI to enhance human capability or let it quietly replace it. The first path creates stronger decision-making and sustained innovation. The second leads to dependency, declining quality, and the slow erosion of institutional knowledge.

Leaders must set the tone. Define clear boundaries for AI, insist on data integrity, and reward employees for using judgment. The businesses that keep human intelligence at the center of technological growth will build systems that adapt, learn, and endure.