The new bounded rationality: why more AI does not mean better management
Long before artificial intelligence made it onto boards of directors' agendas, Herbert A. Simon already saw computers as instruments for understanding and amplifying human intelligence. An economist, psychologist, political scientist and one of the pioneers of AI, Simon received the Nobel Prize in Economics in 1978 for a seemingly simple but revolutionary idea: human beings do not decide as perfectly rational agents.
We decide with incomplete information, scarce time and a limited capacity to process everything that is at stake. We do not see all the alternatives, we do not calculate all the consequences and we rarely choose the best possible solution. In practice, we look for an answer that is good enough to act on.
Simon called this condition “bounded rationality.” The expression is precise because it does not deny human reason; it simply places it back in the real world. Rationality exists, but it operates under constraints: it depends on the quality of the information available, the pressure of time, the complexity of the environment and the judgment of whoever is deciding.
This concept, formulated in the last century, may never have been as relevant as it is now. Artificial intelligence seems, at first glance, to be the most powerful technological answer to the problem Simon described. If our rationality is bounded by insufficient information, processing capacity and time, AI promises to expand precisely those boundaries. It processes more data, compares more alternatives, calculates faster, recognizes patterns invisible to the human eye and simulates scenarios on an unprecedented scale. In theory, AI can make our rationality a little less bounded.
The irony, however, is that by expanding our capacity to decide, it may also reduce our ability to understand. The more we delegate the formulation of answers to the machine, the greater the risk of losing the discipline to formulate good questions. Our rationality is no longer bounded only by a scarcity of data or by slow calculation. It also becomes bounded by dependence, by the opacity of systems that make recommendations without clearly stating their criteria, and by the distance between those who decide and the concrete reality of the work.
This is a critical point for management. AI can identify patterns in volumes of data that escape human perception, predict swings in demand, anticipate bottlenecks, reduce variability, support simulations and speed up analyses. It can help companies react better to market changes, allocate resources more precisely, interpret still-faint signs of change and improve decisions. In complex environments, it can make work flow more smoothly and execution more consistent.
All of this is valuable. But none of it automatically means better management. When applied to poorly designed processes, AI does not eliminate waste; it merely perpetuates and accelerates it. When layered over confusing flows, contradictory goals, irrelevant indicators or merely bureaucratic routines, technology tends to make the problem more sophisticated instead of solving it. The result may look more modern, but it is still waste. Only now with a nice interface, a real-time dashboard and an algorithmic recommendation.
This is where lean thinking gains new relevance.
Lean is not just a repertoire of tools for gaining efficiency. It is a discipline of management and learning. Its strength lies in bringing decisions closer to reality. Before automating, you need to understand the value stream. Before optimizing, you need to know what value is for the customer. Before trusting a model, you need to observe how the work actually happens.
In practice, this means going to the gemba, the real place where value is created and problems appear. It means seeing with your own eyes, understanding the facts in context, listening to the people involved and distinguishing symptoms from causes. This requires the use of PDCA, A3 Thinking, visual management, standardized work and kaizen. Not as administrative rituals, but as mechanisms for thinking better.
If Simon taught us that human rationality is bounded, lean management offers a practical response to that condition. It turns decision-making into learning: it makes problems visible, organizes hypotheses, tests solutions, incorporates experience and develops people in the process. It is, in essence, an architecture of collective rationality.
AI can strengthen this architecture. By cross-referencing information on demand, capacity, quality, cost and delivery, it helps detect anomalies before they turn into failures, anticipate risks, reduce variability, synthesize technical knowledge and speed up analysis cycles. Used well, it expands the organization's ability to perceive deviations, formulate hypotheses and act more precisely.
However, the benefit only appears when technology turns into practical action in the real work. If AI merely produces complex reports that few understand or recommendations that no one can test, it becomes yet another layer of waste: more processing, more noise and little concrete change. To create value, intelligence must simplify the work, not make it murkier.
This is one of the great risks of the AI era: confusing more information with more understanding. Companies may surround themselves with metrics, alerts, forecasts and dashboards and, even so, still fail to see what is essential. The abundance of data creates a kind of digital fog. The manager feels they know more, but understands less. The work seems transparent; reality, however, grows more distant.
The lean logic offers an antidote to this abstraction. It disciplines the organization to ask simple and difficult questions: Is the problem well defined? Was the cause verified in the real work? Was the hypothesis tested? Did the solution reduce waste or just move the problem elsewhere? Did people understand what changed? Was the learning incorporated into the routine? These questions remain indispensable, even when the recommendation comes from a sophisticated algorithm.
There is also the problem of black boxes. Many AI systems generate predictions, classifications and recommendations without sufficiently explaining how they arrived at that result. In an organization, this can weaken respect for people, not rhetorically, but concretely, by reducing individuals' ability to understand, question and improve their own work.
Respecting people does not just mean maintaining cordial relationships, but creating conditions for them to understand the work, take part in improvement, exercise judgment and develop autonomy. When a professional does not understand the logic behind a recommendation, they lose part of their ability to intervene, learn and correct course when something goes wrong. The decision shifts to a system that few understand and many merely execute. This is a management risk, not just a technological one.
In lean management, good automation has always been linked to the concept of jidoka, often translated as automation with a human touch. The idea is not to build systems that keep running despite problems, but mechanisms capable of detecting abnormalities, stopping the flow when necessary and calling on human judgment to address the cause. The machine should raise people's capability, not distance them from the learning process.
This distinction will become increasingly important. AI can support decisions, but not take responsibility for them. It can suggest paths, but not define the ends on its own. It can speed up analysis, but not do away with observation. It can expand cognitive capacity, but not replace the practical wisdom formed through direct contact with the work.
Many AI systems tend to reproduce patterns present in historical data. If these come from inefficient processes, command-and-control cultures, biased decisions or poorly designed routines, the algorithm may perpetuate the past with the look of the future. Statistical sophistication does not guarantee managerial progress; sometimes it merely makes the status quo more convincing.
That is why the strategic question is not: “Where can we apply AI?” That question usually comes too early. The decisive question is: “What human and organizational capability do we want to expand?” Do we need to decide better? Learn faster? Reduce waste? Improve the value stream? Increase safety? Develop people? Bring leadership closer to the real work? Create more value for the customer? Without that clarity, AI becomes a solution in search of a problem. And solutions in search of problems rarely produce good management.
The most fertile relationship between AI and lean management lies exactly at this point. AI expands processing capacity, while lean management protects the capacity to learn. One helps to see patterns; the other demands verifying reality. One speeds up analysis; the other disciplines action. One can produce better answers; the other helps formulate better questions.
This combination is powerful because it avoids two symmetrical mistakes. The first is rejecting AI as a threat to human judgment. That would be shortsighted: technology has real potential to expand the capacity of individuals and organizations to deal with complexity. The second is treating AI as a substitute for managerial thinking. That would be naive: the quality of a decision still depends on how the problem is framed, the quality of the data, the understanding of the context, the clarity of the objectives and the accountability of whoever decides.
Simon would probably view modern AI with fascination. After all, he was one of the first to imagine machines capable of performing tasks once associated with human intelligence. But perhaps he would remind us of something uncomfortable: more computing power does not mean unbounded rationality. All intelligence, human or artificial, still operates within limits.
The challenge for companies is to design management systems in which AI amplifies human intelligence instead of outsourcing it. It is to use algorithms to strengthen observation, experimentation, learning and continuous improvement. It is to make sure technology does not distance leaders from the reality of the work, but helps them see it better.
AI can make our rationality less bounded, but without a management discipline, it may simply make it more dependent. In the end, the question is not whether the machine thinks. It is whether we, with it, come to think better.
