Value creation increasingly depends on the interaction between humans and artificial intelligence. The technology is a genuine participant in knowledge work, yet it cannot operate without humans. Technically, it needs a human with intent to activate it. Ethically, society expects human judgment and accountability in processes in which AI plays a role.
This (economic and ethical) need for the “human in the loop” requires us to redesign roles to establish an effective division of cognitive labor between people and AI. The very expression “human in the loop” poses so many unanswered questions: Which human? From whom do we expect what? In a loop that does what? To solve which problem? Organizations must figure out how humans add value when AI systems are adopted, especially at scale. How a company answers this question will determine whether its AI transformation succeeds or simply incurs an expensive bill for the technology it bought but could not leverage.
Answering those questions is important for the organization to succeed in creating value, but also for the people who need to find new ways of working and adding value. That applies not only to those whose tasks are taken over by AI but also to those who must change the way they do the tasks that remain theirs.
The problem in many organizations is that the leaders don’t have answers to these questions. The uncertainty often leads to anxiety and fear, hampering the AI transformation because people can’t see the benefits and worry about the consequences. Worse still, it results in conscious or unconscious sabotage by the people who should be leading it.
Finding and providing clarity is fundamental in reducing the anxiety that is preventing real value creation and a return on investment in costly AI applications. This makes the human side of AI transformation one of today’s most pressing leadership challenges and responsibilities. As the next stage of value creation relies on effective interaction between AI and humans, we – the people – will be either blockers or enablers. Leaders must find ways to ensure their teams support and actively participate in transformation, rather than hinder it or miss out. The stakes are high. If leaders get this right, they will enable their people to imagine and conduct previously “impossible” tasks with the help of AI: the kind of breakthroughs that can redefine industries and unlock new chapters in economic development. If they fail, their teams and companies will inevitably fall behind, while others survive and progress.
In my research and work with organizations, I use three frameworks to structure this critical conversation about AI transformation: the Value-Data-People framework, an output equation borrowed from economics, and a version of the Stacey matrix. These tools help executives across industries and geographies to deconstruct, map, and better manage the complexity of the AI-human transformation – and to understand how they and their teams can adapt and create value. Here, I take each framework in turn, examine what it reveals about the human experience of AI transformation, and pose vital questions that leaders (including those who lead themselves and not a team) need to ask and answer.
As execution becomes cheaper, the constraint shifts to judgment, and new roles emerge where old ones dissolved.
The Value-Data-People framework organizes AI implementation around three dimensions that organizations must address in sequence.
The first two can be solved with strategy and investment, and require organizations to allocate resources, build infrastructure, and iterate. For this article, let’s assume that these factors are solved, or at least on the way to getting solved.
The third dimension requires greater scrutiny because it involves the internal experience of the people who must adapt. Failing to address it means the technology and data will deliver only a fraction of their potential. The human factor has become increasingly complex as AI tackles sophisticated tasks once reserved for people, a phenomenon referred to as the substitution or enhancement paradigm. Substitution focuses on tasks that AI can perform better or cheaper than humans, while enhancement focuses on tasks that people can do better with AI.
People whose tasks are absorbed by AI and those whose productivity is multiplied by AI share more in common than most leaders realize. Both face a redefinition of their roles and the development of capabilities they were never trained for. And both must answer the existential question: what is my contribution to value creation now? Answering that question will identify the constraints that emerge when AI transformation succeeds, help acquire the skills needed to mitigate those constraints, and adapt hiring and training programs to attract and nurture the right human talent. This ensures that value can be created and captured.
Knowledge workers spend years building professional identities around specific cognitive tasks: financial analysts build models, lawyers research precedents, and engineers write code. When AI begins performing these tasks, it not only raises questions about their future employability but also challenges their sense of identity and self-worth. A team of credit analysts at a bank that has deployed an AI system capable of producing assessments in minutes faces a question that no amount of tool training can answer: if the machine does what the team was known for, what value, if any, does the team bring now?
The most common response is rejection or withdrawal, which is catastrophic for transformation. Professionals disengage, use AI minimally, and find reasons to preserve the old way of working. Leaders who interpret this solely as resistance to technology are misdiagnosing the problem. Resistance is not just about self-preservation; it is a professional identity defense mechanism that will persist until the organization helps people build a new identity they can believe in – one that formal systems reinforce through job descriptions, performance reviews, and promotion criteria.
Building that new identity requires people to answer the question: what does it say about me that I’m using AI in this way to do my job? The answer shapes identity as well as the ability and willingness to engage in adoption. Leaders must identify human-AI interactions that enable people to solve their identity questions to avoid withdrawal and rejection. This often involves understanding where in the value chain we can apply human-only judgment. Successful identity evolution happens when workers see themselves as necessary in making the technology work, can direct the AI toward the right problems, evaluate its outputs with domain expertise, catch the edge cases and contextual nuances it misses, and take responsibility for the decisions that follow. Creating space for this new identity to emerge increases the chances that human-AI interactions will yield value. For credit analysts, for example, this means shifting from producing assessments to ensuring those assessments align with the bank’s risk appetite, regulatory obligations, and client relationships.
A senior credit risk analyst at a European bank spent a decade building the stress-testing models her institution relies on for regulatory reporting. Now, three of the five tasks she was hired to perform have been absorbed by an AI system that delivers in minutes what used to take her days. The fear in her department is obsolescence, but she does not ask whether AI will replace her. She asks what the bank will be missing once the automation has fully settled in. Her answer is simple: there will be no one who can walk into a room with a regulator and defend a model rejection decision in plain language, under pressure, to an institution that distrusts black boxes. She is currently completing a certificate in algorithmic accountability and sitting in on every regulatory examination the bank undergoes. This is something that she can initiate. Leaders play a crucial role in identifying such situations and creating space for people to learn what is needed in the near future.
Professionals need to have faith in the process if they are to redefine their identities. This trust operates in three dimensions. First is trust in the technology: do the professionals believe AI’s outputs are reliable enough to act on? Second is trust in the organization: do employees believe they will be given the time and resources to develop new capabilities, be redeployed rather than discarded if their roles shrink, and be valued for their contribution, even as the nature of that contribution changes? Third is trust in the direct leader: do employees believe their managers understand what they are going through, tell them the truth, and advocate for them? Employees who trust their managers will experiment with AI even when they are uncertain about everything else. Experimentation is where the most valuable discoveries about the potential of AI-human collaboration tend to happen. Leaders who narrate their own experimentation with AI tools build trust in the technology. While openly sharing their own experiments, leaders also signal when AI output deserves scrutiny and when it can be acted on. Leaders who advocate for redeployment and development signal organizational trust. And leaders who tell uncomfortable truths about the transition, consistently and early, earn the trust that makes everything else possible.
Trust is intertwined with psychological safety. In most organizations I have worked with, the dominant culture around AI is performative: people use it to demonstrate compliance with transformation, but they do not use it in ways that genuinely change their work because that carries risks in both directions. If the output is flawed, they look incompetent. If the output is excellent, they look dispensable. The rational response in an environment without psychological safety is minimal engagement, which means minimal value creation.
The starting point is to acknowledge that most organizations are still figuring this out. No established playbook exists for human-AI collaboration; only experiments of varying quality. That admission matters because when leaders present AI transformation as a solved process to be rolled out, rather than a collective learning journey to be navigated, they make it risky for people to say what is not working. Fear, as every educator knows, is the most reliable killer of learning. Psychological safety is not a cultural nicety but an operational prerequisite: without it, people go through the motions of using AI without ever discovering the new forms of value that the collaboration can produce.
One specific hack and practical advice: As a leader, show openly how you are using AI. In team meetings, disclose your conversations with the chat, showcase how you pressure test the answers, how you provide context, and how you use your human agency to decide what to take and what to reject from the AI. This simple gesture will have a tremendous impact in your team.
Am I leading the transition within my own team, recognizing that the IT and technology functions are themselves being reshaped by AI, and that my people face the same identity, trust, and capability challenges as everyone else?
Understanding what makes people willing to collaborate with AI matters because when they experiment and learn, the economic returns are far larger than most organizations capture. But how do we measure these returns and the value that the human-AI collaboration can bring to an organization? And how do we assess human value in the age of AI?
Borrowing from economic theory, let’s say that AI and data enter the organization as a new factor of production alongside capital and human labor. We can then deploy a simplified version of an equation that economists use to explain how factors of production (capital, labor, data, and AI in the case I am illustrating here) create value:
Y stands for the value an organization produces. The rest of the equation says that this value is a function of capital, labor, AI, and data working together. The strategic question for leaders is: with which combination of factors does the organization have the greatest opportunity to create something distinctive? Or put differently, how intensively the organization should use each factor to create a competitive advantage. Given that humans are needed and wanted in the creation of value with AI, this framing allows executives to think about value and substitution in a more granular way.
Let’s take an example. A law firm that uses AI to draft routine contracts is substituting. A lawyer who uses AI to analyze 10 times more case precedents before advising a client is being enhanced. A firm that uses that enhanced lawyer, freed from drafting, to enter a market for complex cross-border advisory work it never had the capacity to serve before is expanding, generating revenue from a capability that did not exist in the organization before AI entered the equation.
Surfacing these three dynamics matters, not least, because they redefine compensation. In any production function, the marginal cost of a factor reflects its marginal contribution to output – in other words, how much a firm needs to invest in that factor to produce one extra unit of a good or service. For human labor, that marginal cost is salary. As AI absorbs substitutional cognitive work, the marginal contribution of professionals in those roles declines, and their market wage falls with it. As AI enhances professionals, the marginal contribution of their judgment rises as it becomes the input that determines whether AI’s output creates or destroys value, thereby exerting upward pressure on compensation. At the expanded frontier, where professionals are generating forms of value the organization has never captured before, the marginal contribution of their capability has no historical benchmark, and their compensation will reflect that novelty.
Organizations that retain legacy compensation structures will systematically misprice the inputs that matter most to value creation. The equation under discussion gives executives a framework for that repricing by making visible which factors are gaining or losing marginal contribution, and by connecting compensation decisions to the underlying production logic rather than historical precedent and internal politics. It allows leaders to ask, with clarity, whether they are allocating resources across capital, labor, AI, and data in a way that reflects where value is being created. Executives can also use it to read their competitors and their industry with the same logic, asking where rivals are investing across the four factors, where the industry is substituting or enhancing, and where the expanded frontier is most likely to emerge.
Developed by the management professor and organizational theorist Ralph D Stacey, the Stacey matrix enables executives to determine how expansion will specifically affect their organizations. It categorizes tasks or problems to be solved along two dimensions: certainty of outcomes and agreement among stakeholders. Mapping the tasks a person or team conducts in this way allows leaders to identify the evolution required to achieve the expansion and increase in value creation that AI, in combination with humans, makes possible.
Simple tasks (high certainty, high agreement) face deep substitution. They include data entry, standard reporting, routine compliance, and repetitive processing. Stanford’s Digital Economy Lab has found that entry-level hiring in AI-exposed jobs has dropped 13% since the proliferation of large language models. The value case for automation is compelling, but the people challenge is immediate and consequential for the organization.
How people who lose their roles are treated becomes the most powerful signal to the rest of the workforce about whether they can trust the organization. As execution becomes cheaper, the constraint shifts to judgment, and new roles emerge where old ones dissolved. The analyst who once rebuilt the same weekly dashboard now faces 10 times the volume of machine-generated analyses, most of which lack provenance or argument. The emerging role curates which questions deserve a model, audits its logic, and translates outputs into decisions a board can defend, which is harder work requiring a different person than the one let go. For this to happen, training is needed. And training is expensive.
The value of AI depends on the quality of the human judgment directing it. The people side is the variable most worth developing.
Complicated tasks (moderate certainty, moderate agreement) face partial substitution and significant enhancement. They include financial modeling, engineering analysis, legal research, and diagnostic imaging. AI absorbs the production subtasks (the research, computation, drafting, and initial analysis) while professionals retain value in judgment, problem framing, and exception handling. Teams shrink but become more capable, producing results that neither humans nor AI could achieve alone.
The people challenge is to redesign roles around judgment rather than production, and to rebuild team structures around fewer and more skilled professionals working in collaboration with AI. Leaders need to shift performance metrics from output volume to the quality of judgment applied and to reimagine career paths for a world where the traditional junior-to-senior progression no longer holds.
This creates one of the biggest challenges in AI transformation: if juniors no longer perform the entry-level tasks that build expertise, how do they develop the judgment that collaboration depends on? One possible answer is a new kind of apprenticeship, where juniors learn by evaluating AI-generated work, identifying its errors, and building the domain-specific judgment that allows them to direct AI effectively. This solution, honestly, is also expensive.
Complex tasks (low certainty, low agreement) are where human primacy remains. They include strategic decision-making, organizational change, creative direction, negotiation, ethical reasoning, the ability to build trust across relationships, and the willingness to take responsibility for decisions under uncertainty. AI can provide inputs, research, and analysis, but current systems struggle with the edge cases, the relational dynamics, and the genuine ambiguity that define this zone. Developing human capability here should be a strategic priority for any organization seeking to capture the full value of human-AI collaboration.
Most organizations underinvest in developing these capabilities because they are hard to measure and teach, yet investment in this area will determine competitive advantage more than any decision about technology. Organizations that focus exclusively on tool adoption will find that their people can operate AI efficiently without being able to exercise the judgment needed to turn its outputs into good decisions.
Beyond complexity lie tasks that were previously impossible to accomplish because the cognitive load exceeded human capacity. AI moves the boundary of this zone, creating new categories of potential value. The challenge lies in finding and developing professionals who can see that a problem has become solvable through AI, formulate it for a human-AI team, and navigate a domain with no precedent and no established method. Leaders need to identify and protect these people, provide resources, and give them the latitude to explore, with accountability and psychological safety. It is these talented individuals who will discover the forms of value that the organization does not yet even know it can capture. Interestingly, the ones most capable of doing this seem to be relatively young – the very people who are allegedly most under threat.
The technology will continue to advance and the economics will keep shifting. Some will argue that the endpoint of that trajectory is a production function in which human labor is no longer a meaningful variable. The argument deserves to be taken seriously, and it may prove correct in domains and over timeframes that we cannot yet specify.
What leaders must act on, however, is the reality they face today. We are in a phase in which the value of AI depends on the quality of the human judgment directing it. Organizations that treat the people side of this equation as a constraint to be minimized will systematically underperform the ones that treat it as the variable most worth developing. The responsibility lies on the shoulders of the leaders – both those managing teams and those “only” leading themselves – who need to decide how much autonomy they give AI and how much agency they still want to exercise.
AI is developing a “model-year” pricing dynamic, much like cars, with yesterday’s capabilities rapidly becoming commoditized, while this year’s frontier models stay expensive. Michael Watkins explains why and what this means for...
As AI spreads through the enterprise, companies are losing something that few have thought to protect: the capacity to judge what their systems are doing and to change course when they need...
AI can optimize processes, but true leadership requires human judgment, says Shelley Zalis. Success depends on integrating machine intelligence with ‘soft skills’ like empathy and intention....
Audio available
Facing a fundamental redefinition of their purpose in organizations, today’s CIOs need to pivot quickly. ...
Explore first person business intelligence from top minds curated for a global executive audience