AI Leadership Is Not Tool Training

It Is the Redesign of Executive Judgment, Organizational Capacity, and Human–Machine Work

8/24/202618 min read

Author: Trang Phan

Introduction — The First Generation of AI Training Taught Leaders Which Tools to Use; the Next Must Teach Them How to Lead a Different Kind of Organization

Most executive AI training is currently too shallow for the transformation underway. A typical leadership programme introduces ChatGPT, Microsoft Copilot, Claude, Perplexity, Power BI, meeting summarizers, presentation generators, workflow automation tools, and perhaps a small library of prompts. Leaders learn how to summarize a document, draft an email, prepare a board presentation, research a competitor, automate meeting notes, or extract insights from spreadsheets. These are useful skills. They reduce friction, increase personal productivity, and help executives become comfortable interacting with generative systems. That is an excellent entry point. It is no longer sufficient as the end point.

The next stage of AI leadership is not about teaching senior executives how to use ten applications. It is about teaching them to redesign the allocation of cognition inside the enterprise. For more than a century, companies organized work around a relatively stable assumption: human cognitive capacity was scarce, expensive, slow to scale, and tied to individual employees. Organizations therefore created hierarchies to distribute information and decision rights. Analysts gathered information. Managers synthesized it. Senior managers escalated important issues. Executives decided. Administrative staff coordinated the process.

AI changes this foundational constraint. Research from Microsoft describes a rapidly emerging environment in which intelligence can increasingly be deployed on demand through AI systems and agents. In its 2025 Work Trend Index, 82% of surveyed leaders said that year was pivotal for rethinking major aspects of strategy and operations, while 46% said their organizations were already using agents to automate workflows or processes. The same research found that 80% of workers reported lacking sufficient time or energy to complete their work, suggesting that organizations face not simply a technology opportunity but a capacity-allocation problem. ()

McKinsey reached a related conclusion from another direction. Its 2025 workplace research found that almost all companies were investing in AI, yet only 1% considered themselves mature in deployment. McKinsey's diagnosis was striking: employees were often more ready to use AI than leaders assumed, while organizational leadership and operating-model redesign were becoming the larger constraint. () The implication is profound: the leadership challenge is moving from AI adoption to organizational redesign.

The executive of the AI era therefore needs more than technological literacy. They need cognitive architecture literacy. They must understand which work should remain human, which should become machine-assisted, which should become machine-operated, which should require dual control, which decisions can be delegated, which must remain accountable to a named human, how AI-produced evidence should be challenged, how organizational memory should work, how agents should receive authority, and how people should develop when execution itself is increasingly automated. This is not IT training. It is leadership training for a different organizational form.

1. The First Principle — AI Does Not Remove the Need for Leadership; It Moves Leadership Upward

One of the most persistent misconceptions surrounding AI is that increasing machine capability necessarily reduces the need for human leadership. The opposite is more likely in consequential environments. As AI performs more analysis, generation, coordination, and execution, leadership becomes increasingly responsible for what exists above those activities: What objective is being optimized? What evidence is admissible? Which risks are acceptable? Who has authority? When must a human intervene? What values cannot be traded for efficiency? Which failure modes are tolerable? How should competing objectives be prioritized? What happens when the AI recommendation is technically plausible but strategically wrong?

These are governance questions. Technology cannot resolve them independently because they are not merely computational questions. They involve authority, accountability, institutional purpose, and consequence. A leader who delegates execution without redefining responsibility does not create an AI-enabled organization. They create an accountability gap. The leader's role becomes more, not less, important as automation increases, because the consequences of poorly governed automation are more severe than the consequences of poorly governed manual work.

2. The Second Principle — Leaders Should Stop Thinking of AI as Software and Start Thinking of It as Cognitive Capacity

Traditional enterprise software performs predefined operations. CRM systems store customers. ERP systems coordinate resources. Accounting platforms record transactions. Workflow applications route tasks. Generative AI and agents introduce a different category of capability. They can interpret ambiguity, generate alternatives, search information, summarize evidence, draft analyses, critique arguments, plan sequences, use tools, and adapt outputs to context. In sufficiently bounded environments, they can perform significant portions of knowledge work. The strategic resource is therefore not merely software access. It is additional cognitive capacity.

Microsoft's 2025 research describes this transition through the concept of "intelligence on tap" and argues that organizations are beginning to treat AI agents as a new source of workforce capacity rather than simply another productivity application. () This should change how executives budget and organize. Historically, a CEO confronting analytical overload might request additional analysts. A commercial leader might request additional sales operations staff. A strategy director might request consultants. A finance function might hire more junior analysts. Some future capacity decisions will instead ask: Should this requirement be solved by another employee? An AI assistant? A specialized agent? A human-agent team? An automated workflow? A redesigned process that eliminates the requirement entirely? This is a fundamentally different management discipline.

3. The Third Principle — The Executive's Most Important AI Skill Is Not Prompting; It Is Problem Decomposition

Prompting is not primarily a writing skill. At high levels, it is a problem-structuring skill. Weak executives give AI vague tasks. Strong AI-enabled executives define the objective, the relevant context, the decision boundary, the evidence required, the constraints, the alternatives, the uncertainty, the expected output, and the conditions under which the conclusion should be rejected. Compare two instructions. The first says: "Analyze our declining sales and suggest a strategy." The second says: "Separate the decline into market contraction, share loss, price effects, channel effects, customer churn, and product-mix effects. Identify which hypotheses are supported by supplied evidence, which remain untested, what additional data would discriminate among them, and which interventions remain reversible while uncertainty is high."

The second instruction is not merely a better prompt. It reflects better executive thinking. AI therefore exposes an uncomfortable truth. Poorly structured thinking becomes more visible when a machine executes it quickly. A vague leader with AI can generate vague work faster. A precise leader with AI can increase analytical leverage dramatically. The quality of the instruction determines the quality of the output, and the quality of the instruction reflects the quality of the thinking behind it.

4. AI Training Should Begin with Decision Quality, Not Tool Demonstration

Executives are rarely valuable because they write the fastest email or summarize the most reports. Their economic value comes disproportionately from a relatively small number of consequential decisions: capital allocation, market entry, senior hiring, pricing, acquisitions, product portfolio, risk acceptance, technology investment, organizational redesign, crisis response, and strategic prioritization. Leadership AI training should therefore begin with the executive decision cycle: What do I need to know? What can AI discover? What evidence must remain independently verified? What alternatives am I ignoring? What would falsify the current recommendation? Which assumptions drive the decision? What is reversible? What is irreversible? Who owns the consequence?

AI should become part of the decision architecture—not merely the document-production architecture. The executive's use of AI should be judged not by how many documents they generate but by whether the quality of their decisions improves. This shifts the focus from productivity to judgment, from speed to wisdom, and from automation to augmentation.

5. The New Executive Operating Model Has Five Levels

Level One — AI as Assistant

AI helps a leader summarize reports, draft communication, prepare meeting notes, research markets, generate questions, translate documents, analyze simple information, and organize personal work. The human retains virtually all decision-making and execution responsibility. This stage is valuable because it builds familiarity and reduces resistance. But the productivity ceiling is limited. The executive is essentially performing the same work faster.

Level Two — AI as Thought Partner

At the second level, the executive stops using AI merely for production. They begin using it to challenge thinking. The system is asked to construct competing hypotheses, identify missing evidence, argue the opposite case, simulate stakeholder perspectives, identify second-order consequences, stress-test plans, compare scenarios, and expose hidden assumptions. Microsoft's 2025 survey found that 46% of respondents already described AI as a thought partner rather than solely a command-based tool. () This is a major transition. AI moves from doing work for the executive to improving how the executive thinks about the work.

Level Three — AI as Delegated Analyst

The third level involves bounded delegation. Instead of asking AI isolated questions, leaders assign analytical missions. Examples include monitoring competitors, reviewing financial variance, tracking policy developments, preparing customer-risk briefings, conducting initial investment screening, monitoring operational anomalies, or synthesizing weekly strategic intelligence. The human still controls important decisions, but information preparation becomes increasingly machine-operated. This can dramatically reduce management latency.

Level Four — Human-Agent Teams

At the fourth level, AI is no longer attached to individual executives. It becomes embedded in organizational workflows. A commercial team may use agents for customer research, proposal preparation, and pipeline analysis. Finance may use agents for reconciliation, variance explanation, and forecast preparation. Operations may use agents for anomaly detection and scheduling. Strategy teams may use AI to monitor external signals continuously. Microsoft reports that leaders increasingly expect employees to train and manage agents, with 41% expecting teams to be training agents and 36% managing them within five years. () At this stage, leadership must manage not only people but systems of people and digital workers.

Level Five — Human-Led, AI-Operated Systems

The fifth level is the most consequential. Humans retain purpose, authority, ethical responsibility, and strategic judgment. Large parts of operational execution become machine-managed. This does not mean removing people indiscriminately. It means deliberately separating purpose from execution, judgment from processing, authority from capability, and exception handling from routine work. This operating model requires far more sophisticated leadership than conventional automation.

6. The Future Leader Becomes an Architect of Human–AI Work

Leadership has historically meant allocating human resources. Increasingly, it will involve allocating human cognition and machine cognition together. A leader must learn to ask: Where does AI outperform people? Where do people outperform AI? Where does combining them outperform either alone? Where must a human remain legally or ethically accountable? Where does a customer require human interaction? Where does AI remove low-value work that prevents humans from performing higher-value work? Where does automation create new fragility? This cannot be solved through a company-wide mandate such as "use AI everywhere." Different work requires different architectures.

A repetitive internal reporting process may become largely automated. A high-stakes investment committee should not. A customer-service workflow may be heavily agent-assisted while retaining human escalation. An engineering workflow may automate code generation while retaining testing, architecture, and release governance. The leader therefore becomes a designer of human–machine boundaries. This requires understanding not just what AI can do but what it should do, and not just what humans can do but what humans should continue to do.

7. The Most Important New Leadership Question Is: What Should Humans Stop Doing?

Most AI programmes begin by asking: Where can we use AI? That question encourages scattered experimentation. The stronger question is: What work no longer deserves human attention? This forces the organization to examine its operating model. How much leadership time is spent retrieving information? Reformatting information? Reconciling reports? Preparing repetitive presentations? Following up on actions? Rewriting existing content? Searching documents? Preparing status updates? Coordinating meetings whose primary purpose is information transfer?

These activities once existed partly because organizational cognition was expensive. AI changes the economics. If machines can absorb significant portions of these activities, leaders should not merely use the recovered time to perform more administration. They should move attention upward toward strategy, customers, talent, innovation, risk, culture, institutional design, and difficult judgment. The question is not what AI can do. The question is what humans should stop doing because machines can now do it better, faster, or more consistently.

8. The Board Should Govern AI as a Workforce and Operating-Model Issue

Boards frequently discuss AI as technology investment, cybersecurity, digital transformation, or innovation. Those categories are now insufficient. AI increasingly affects workforce planning, organizational structure, capital allocation, customer experience, competitive strategy, risk, management information, talent development, and executive succession. The board therefore requires a new question set. What percentage of core workflows has been redesigned rather than merely augmented? Where is AI creating measurable economic value? Where is automation reducing organizational learning? What decisions are being machine-assisted? Which are machine-executed? Where does human accountability remain explicit? What proprietary data or workflow advantages are being created? What capabilities are competitors developing faster? Which roles are changing? Which new roles are emerging? How are junior employees developing expertise when AI performs work previously used for apprenticeship?

This final question is particularly important. Many professional organizations develop senior people through repetitive junior work. Analysts build financial models. Associates conduct research. Engineers debug code. Junior lawyers review documents. Consultants synthesize data. Managers learn through operating experience. If AI increasingly performs the entry-level work, organizations may gain immediate productivity while weakening the mechanism by which humans develop expertise.

9. AI Creates a Leadership Pipeline Problem

This creates a paradox: AI can automate the work that trained people to supervise AI. Leadership training must therefore redesign apprenticeship. Junior employees may need earlier exposure to judgment while simultaneously learning verification, problem decomposition, system design, and AI supervision. Organizations that automate without rebuilding learning pathways may become highly efficient in the short term and strategically hollow in the long term. The leadership pipeline requires deliberate redesign to ensure that the next generation of leaders develops the judgment that can only be acquired through experience, even as the nature of that experience changes.

10. The Executive Must Learn to Manage AI Error Differently from Human Error

Human error and AI error have different distributions. Humans become tired. They forget. They operate slowly. They may possess narrow knowledge. AI can process information rapidly and consistently. But it can also generate plausible falsehoods, misread context, follow an incorrect objective at scale, or repeat the same flawed logic across thousands of decisions. This means leadership must distinguish local error from systemic error. A human analyst making one mistake affects one analysis. An automated agent with the same mistake embedded in its workflow can affect every analysis.

Therefore increasing automation requires stronger validation architecture. The executive rule should be: the more scalable the AI decision, the stronger the control around the decision. Leaders must understand that AI errors are not random—they are systematic. A flawed prompt, a biased dataset, or a mis-specified objective can propagate through thousands of decisions before being detected. This requires different oversight mechanisms than traditional quality control.

11. AI Literacy Must Include Knowing When Not to Trust AI

Modern AI leadership requires a minimum understanding of failure. Executives should understand at least five ideas. First, fluent language does not guarantee factual accuracy. Second, AI systems can reproduce biases contained in data and evaluation processes. Third, access to confidential information creates privacy and security implications. Fourth, AI can optimize the wrong objective extremely efficiently. Fifth, automation can transform a small design mistake into a large operational mistake. This is not engineering education. It is management literacy. A CEO does not need to design an aircraft engine to run an airline. But a CEO who does not understand the implications of maintenance, safety, capacity, and reliability cannot responsibly run the airline. AI will become similar.

12. The Leader Must Separate Capability from Authority

One of the most important principles for agentic AI is simple: just because an AI can perform an action does not mean it should be permitted to perform it. AI may be capable of sending an email, approving a refund, changing pricing, moving money, modifying production code, creating a supplier order, publishing communication, or making staffing recommendations. Capability is technical. Authority is institutional. The two must remain separate. A mature organization grants AI authority according to risk, reversibility, confidence, impact, data sensitivity, and legal accountability. This principle will become fundamental as AI moves from content generation toward action. The leader's role is to define and enforce the boundary between capability and authority.

13. AI Transformation Should Begin with Workflow Redesign, Not Headcount Reduction

Cost reduction will inevitably be one motivation for enterprise AI. But beginning with the question "How many people can AI replace?" usually produces inferior transformation. It frames employees as cost units rather than sources of knowledge required to redesign the system. The better sequence is: understand the work, identify repetitive cognitive burden, capture domain expertise, redesign the workflow, automate bounded tasks, measure quality, reallocate human capacity, then redesign roles. This preserves knowledge during transition. Microsoft's research illustrates the tension: while 33% of surveyed leaders were considering headcount reductions, 78% were also considering hiring for new AI-related roles. () This is not simply substitution—it is workforce recomposition.

14. AI Training Should Therefore Have Three Tracks

Personal Augmentation

Every leader should become capable of using AI as a high-quality personal cognitive assistant. This includes research, writing, analysis, meeting preparation, communication, document review, and idea development. This remains the foundation.

Decision Augmentation

Leaders must then learn to use AI to construct scenarios, surface assumptions, compare alternatives, identify missing information, challenge proposals, and improve the quality of strategic decisions. This is where the greatest leadership leverage begins.

Organizational Augmentation

Finally, leaders must learn to redesign teams, workflows, decision rights, and capability systems around AI. This is the step most current training programmes omit. Yet it is the step that determines whether AI remains a productivity tool or becomes an enterprise capability.

15. The New Leadership Competency Model

The World Economic Forum's 2025 Future of Jobs research finds that technology-related skills such as AI and big data are among the fastest-growing areas of demand, while analytical thinking remains a leading core skill and resilience, flexibility, leadership, and social influence continue to matter. Nearly 40% of workers' core skills are expected to change by 2030. () This reinforces a crucial point: AI does not reduce the value of human leadership capabilities. It increases the value of higher-order leadership capabilities.

The leader of the next decade requires AI literacy without technology worship, data literacy without blind quantification, strategic thinking, systems thinking, critical reasoning, ethical judgment, organizational design, change leadership, curiosity, learning agility, and the ability to distinguish reversible experimentation from irreversible consequence. The executive advantage will come from combining human judgment with machine leverage rather than maximizing either alone.

16. The Leader's Role Shifts from Information Bottleneck to System Orchestrator

Historically, senior leaders often accumulated power by controlling information. Reports moved upward. Decisions moved downward. AI weakens this structure because analytical capability can become more distributed. A frontline employee may have immediate access to sophisticated analysis. A manager can generate scenarios without waiting for a strategy team. A salesperson can analyze customer information independently. A factory supervisor can query operational data directly. This changes leadership. The leader should no longer attempt to remain the smartest information processor in the room. That position becomes impossible. Their role becomes: set direction, create constraints, allocate authority, resolve conflicts, develop people, protect institutional integrity, and ensure that distributed intelligence produces coherent action. This is leadership as orchestration.

17. AI Should Reduce Management Latency

Large organizations frequently suffer not because they lack information but because information travels too slowly. A problem occurs Monday. The report is prepared Wednesday. Management reviews it Friday. A committee discusses it next Tuesday. Leadership approves action the following week. By then, the underlying condition may have changed. AI can compress this cycle dramatically. But only if the organization changes the workflow around the technology. Automating the report while preserving every approval layer yields limited value. The deeper question is: how much decision latency exists because of information-processing constraints that no longer need to exist? This may become one of the largest productivity opportunities in complex organizations.

18. AI Also Creates the Risk of Decision Velocity Exceeding Governance Capacity

Speed is not inherently good. A bad decision made slowly causes damage. A bad automated decision executed thousands of times per second causes more. Therefore AI transformation must balance decision velocity against governance capacity. High-volume, low-consequence decisions can be heavily automated. High-consequence, irreversible decisions require stronger review. The operating principle should be: increase autonomy as consequence decreases and evidence quality increases. This creates a rational architecture for delegation.

19. The Best AI-Enabled Company Will Learn Faster Than Its Competitors

Competitive advantage increasingly depends on learning velocity. Two companies may possess the same AI models. One uses them to generate more presentations. The other uses them to accelerate the entire learning loop: observe, hypothesize, experiment, measure, correct, standardize, repeat. The second company compounds knowledge. That difference eventually becomes enormous. The real AI moat is therefore not simply technology access. It is the speed and quality with which the organization converts experience into improved future action. The organization that learns faster will outcompete the organization that merely automates faster.

20. Leaders Must Create Safe Space for Employee Experimentation

McKinsey found employees were using generative AI more extensively than many executives realized; C-suite leaders estimated far fewer employees used AI intensively than employees themselves reported. () This means experimentation will happen whether leadership formally designs it or not. The strategic choice is between governed experimentation and shadow experimentation. A strong organization creates safe sandboxes, provides approved tools, defines confidential-data rules, creates examples, shares successful workflows, provides expert support, reviews high-impact automations, and rewards useful experimentation. The objective is not to suppress bottom-up innovation. It is to convert it into institutional capability.

21. Executive Training Should Be Experiential

AI cannot be learned through lectures alone. Executives need to work on real decisions. A serious leadership programme should require each participant to bring one repetitive personal workflow, one difficult strategic decision, one information bottleneck, one team process, and one business problem where AI might materially change economics. They should then redesign those cases during training. This produces capability rather than awareness. The training outcome should not be "I understand what generative AI is." It should be "I have changed how I work, changed one decision process, and redesigned one organizational workflow." That is a much higher standard.

22. The 7-Day Introduction Should Become a 90-Day Leadership Transformation

For executive development, the first week of personal fluency should become the first phase of a longer transformation. Days 1–7 focus on personal fluency: each leader uses AI personally every day on real work. The objective is familiarity. Days 8–30 focus on cognitive leverage: leaders use AI for scenario development, critique, analysis, and decision preparation. The objective is improved thinking. Days 31–60 focus on workflow redesign: each leader selects one team process and decomposes it into human-only, AI-assisted, AI-executed, and human-governed components. The objective is operating-model redesign. Days 61–90 focus on scale and governance: the organization compares experiments, establishes standards, defines governance, and scales high-value workflows. The objective is institutional capability. This progression avoids the two common extremes: months of strategy without practical use or uncontrolled experimentation without architecture.

23. The Board Should Expect Five Outcomes from AI Leadership Training

The first outcome is time released. Executives should demonstrably spend less time processing low-value information. The second is better decision quality. Major decisions should contain stronger evidence, clearer alternatives, and explicit assumptions. The third is workflow redesign. Teams should eliminate unnecessary work rather than simply accelerate existing work. The fourth is organizational learning. Successful patterns should become reusable. Failures should generate lessons rather than disappear. The fifth is governed autonomy. AI should perform increasingly valuable work without creating uncontrolled authority. Training that produces only tool familiarity has not reached the leadership level.

24. The Biggest Leadership Failure Will Be Delegating Thinking Before Learning How to Think with AI

AI makes cognitive outsourcing extremely easy. That creates a new risk. Executives may ask AI what our strategy should be, whether we should enter this market, whether we should acquire this company, which employee we should promote, or which investment we should make. The system can produce confident answers. But leadership cannot outsource responsibility merely because analysis becomes easier. The correct use of AI is not "think instead of me." It is "increase the range, speed, and rigor of the thinking for which I remain responsible." That distinction should become foundational executive doctrine.

25. The Second Biggest Failure Will Be Using AI to Preserve Bad Management

An inefficient organization can use AI to produce inefficient reports faster. A bureaucratic company can automate bureaucracy. A meeting-heavy organization can use AI to summarize unnecessary meetings. A badly designed approval process can become digitally accelerated. None of these creates transformation. AI magnifies structure. Good structure becomes more capable. Bad structure becomes faster. Therefore before automating a process, leadership should ask: should this process exist at all? This one question can prevent enormous automation waste.

26. The Third Failure Will Be Confusing AI Adoption with AI Advantage

Every major organization will eventually have access to capable AI. Access therefore cannot remain the moat. Competitive advantage will come from proprietary context, better workflows, faster learning, better organizational memory, stronger human talent, trusted customer relationships, integrated data, and superior governance. The organization that merely buys AI licenses will not necessarily outperform. The organization that redesigns itself around intelligence probably will. The question is not whether we have access to AI. The question is whether we have redesigned our organization to leverage it.

27. The New Executive Question Is Not "Do We Have an AI Strategy?"

A better question is: where is intelligence scarce in our current operating model, and what happens when that scarcity disappears? That question is much more disruptive. It forces leadership to reconsider team size, management layers, decision processes, knowledge access, organizational boundaries, outsourcing, consulting, shared services, training, and even the purpose of certain roles. AI strategy then becomes business strategy. The question of AI is not a separate question. It is the question of how work gets done, how decisions get made, and how organizations learn.

Conclusion — AI Leadership Is the Discipline of Deciding What Humans Should Continue to Own

AI leadership begins with tools but does not end there. The evidence now suggests that organizations are entering the next stage. AI adoption is spreading faster than many executives expected. McKinsey reports that almost every surveyed organization is investing while very few regard themselves as mature. Microsoft finds leaders increasingly redesigning workflows around agents and anticipating that employees will train and manage digital workers. The World Economic Forum expects significant changes in core workforce skills while simultaneously emphasizing continued demand for analytical thinking, resilience, flexibility, and leadership. () The message for boards and executives is therefore not that AI will replace leadership. It is that AI will expose weak leadership faster.

Leaders who cannot define problems clearly will automate ambiguity. Leaders who cannot prioritize will generate more options without making better choices. Leaders who distrust employees will create centralized AI bureaucracy. Leaders who chase cost reduction alone may automate away the learning systems that produce future leaders. Leaders who confuse capability with authority will create governance risk. Leaders who understand the transition can do something very different. They can remove cognitive waste. Expand analytical capacity. Shorten decision latency. Preserve institutional memory. Increase experimentation. Redesign roles. Create human-agent teams. Move people away from repetitive execution toward judgment, creativity, relationships, and system improvement. And ultimately build organizations in which intelligence is no longer trapped inside hierarchy.

That is the deeper transformation. The industrial era taught leaders to allocate capital. The information era taught them to allocate information. The AI era will require them to allocate intelligence. Not every problem should go to a person. Not every problem should go to a machine. Not every decision should be automated. Not every human activity should be preserved simply because humans historically performed it. The central discipline of leadership will increasingly become determining the right boundary.

What should machines do? What should humans do better because machines exist? What decisions must humans continue to own regardless of machine capability? And most importantly: what kind of organization becomes possible when intelligence is no longer the scarce resource—but judgment, trust, purpose, and accountability still are? The leaders who answer those questions well will not simply use AI more effectively. They will build a different class of organization.

References

[1]: https://blogs.microsoft.com/blog/2025/04/23/the-2025-annual-work-trend-index-the-frontier-firm-is-born/ "The 2025 Annual Work Trend Index: The Frontier Firm is born - The Official Microsoft Blog"

[2]: https://www.mckinsey.com/featured-insights/charts/leaders-underestimate-employees-ai-use "Leaders underestimate employees' AI use"

[3]: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-product-and-services/ai/pdf/executive-summary-work-trend-index-annual-report.pdf "Executive Summary: 2025 Work Trend Index Annual Report"

[4]: https://www.weforum.org/stories/2025/01/future-of-jobs-report-2025-whats-shaping-the-future-of-the-global-workforce/ "Future of Jobs Report 2025: What's shaping the future of the global workforce? | World Economic Forum"

[5]: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work/ "AI in the workplace: A report for 2025 | McKinsey"