Human Intelligence and AI in Creating and Shelving Possibilities

Quick Summary

Creating and shelving are a single strategic responsibility. People and AI participate in inquiry and testing. Human capital and social capital enable realization and learning. When we shelve a possibility, we release resources and preserve the learning.

Managing business in an age of uncertainty requires the ability to create possibilities. This ability is built through the joint work of people, AI and partners. We will now examine how to organize it over time.

The company needs to build the ability to create possibilities that fulfill its purpose, and to shelve possibilities that no longer justify continued work. Both actions require thinking, knowledge and the ability to work together. Human intelligence, artificial intelligence, human capital and social capital need to be integrated into this process.

In Chapter 4, we clarified what kind of company we want to build and the foundational principle that will guide our decisions. In Chapter 5, we examined how to measure the creation of economic value. We will now examine how to turn the principle into ongoing organizational work. How to create possibilities, test them and decide where to invest. How to shelve a possibility while preserving the knowledge produced by testing it.

I see creation and shelving as a single strategic responsibility. Creation expands the company’s courses of action. Shelving allows resources to be concentrated on possibilities worth advancing. The organization must be able to do both without losing what it has learned.

Rapid change in AI adds an important dimension to the process. A possibility that was impractical may become practical. An advantage that looked promising may disappear. We will therefore reexamine possibilities and the ways of realizing them, in light of the foundational principle and changes in reality.

The foundational principle guides both the search and what we give up

Purpose defines what we seek to achieve through the company. The foundational principle defines what will guide our choice of how to act. Possibilities are the paths we create to put that choice into practice.

Recall the interpretation I proposed of Dani Ziv’s approach at Plasan. The purpose was to protect people operating in dangerous environments. The principle connected protection with the ability to perform the mission. From it followed the need to connect knowledge of materials with an understanding of the vehicle. The search for possibilities gained a clear direction and clear conditions.

The same principle can also justify shelving a possibility. A solution that adds protection but unacceptably harms performance requires a change or a stop, even if the achievement in armor materials is impressive. The decision follows from the outcome the company is committed to creating.

In my reading of Jensen Huang, the search begins with a bottleneck that prevents an important problem from being solved through computing. Capabilities and partnerships are then built that expand what is possible. Here, too, we need to test whether the bottleneck still exists and whether the company’s contribution is still needed to overcome it.

Before examining a new technology, we will ask what it changes for our purpose. What possibility does it open? Which existing possibility does it strengthen or weaken? And what does it allow us to stop doing?

Commitment to a principle does not exempt us from testing the assumptions it rests on. Purpose is our choice. The claim about the bottleneck is a claim about reality. If reality disproves it, we must return to the foundational inquiry. An investment cannot be justified solely because it once fitted our chosen direction.

The four components within the same work

Human intelligence is the capacity for thought that we exercise. Through it, we interpret situations, ask questions, imagine and examine. It operates through experience, goals and social context, and is also influenced by habits, emotions and authority relationships.

Human capital is the knowledge, skills and experience people have accumulated. Human intelligence puts it to use and develops it. These concepts partly overlap; they are not two assets to be added together in accounting. In creating possibilities, professional experience helps identify what is missing, what is possible and what needs testing before a commitment is made.

AI provides technological capabilities for analysis, proposing alternatives, simulation and execution, depending on the system and conditions of use. It can help us broaden the search and examine competing explanations. Its capability limits change and require testing in practice. A convincing result alone is not a sufficient basis for investment or shelving.

Social capital resides in relationships among people and organizations. It includes trust, norms of sharing and the ability to coordinate action. It enables an expert to voice doubt and a team to disagree with management. It also enables knowledge, resources and customers to be connected through partners. The number of contacts does not demonstrate an ability to act with them.

Integration occurs when the four components change a decision or an action. For example, an employee reveals a need management did not know about. A system helps develop different responses. An expert checks feasibility, and a partner enables an experiment with users. The result shows which possibility deserves to be advanced and which should be shelved.

Management is responsible for designing this work. It must give people the time, information and authority to contribute, and establish who is accountable for the decision and its consequences.

Create possibilities with a path to realization

We will present the team and the AI with the purpose and principle. Clarify which assumptions about reality need testing. Ask for different ways to realize them, including changes to the product, the business model, the capability we build or the division of work with partners.

We will also look for a possibility made available by a new capability. If the cost of an activity has fallen sharply, who can now use it? If work becomes faster, what service can be offered? If fields of knowledge can be connected, what bottleneck can be removed? This brings technological change into the search through the foundational principle.

I suggest maintaining a register of the organization’s possibilities. For each, we will document whom it is for and what it is intended to make possible. Explain how it serves the principle. Define the capability that must be built and who needs to participate. Clarify what each participant will receive and why they would be willing to invest. Appoint someone to lead its examination and define the decisive assumption and the next experiment.

AI can help develop the alternatives, but people who know the organization must supply the context. A production line worker may identify a constraint absent from the report. A service employee may identify a need the customer has not articulated. A partner may point to an approach we cannot carry out alone.

A study of innovation at Procter & Gamble, published in 2026, found that working with AI helped connect technical and commercial perspectives. Participants working alone with it achieved performance similar to that of teams working without AI. [1] The finding concerns the innovation tasks examined. It justifies examining ways to connect knowledge, and does not prove that sustained human partnership can be replaced.

We will also preserve independent thinking before encountering the system’s suggestions. Each participant will first formulate their own understanding of the possibility. We will then compare the proposals and develop them together. In a story writing experiment, AI assistance improved assessments of the stories’ creativity but made the stories more similar to one another. [2] We cannot infer that all strategic work will produce the same outcome. My conclusion is that we need to test whether the assistance expands our range of possibilities.

Test a possibility and choose where to invest

Before a significant investment, we will define what evidence the decision lacks. Examine the assumption whose failure would undermine the proposed approach. It may be an engineering assumption. It may concern a customer’s willingness to pay. Everything may depend on a partner’s willingness to participate. The experiment must test that condition. Even after a good result, we do not know what will happen in the future.

We will divide the work according to demonstrated ability in the task. Establish who proposes, who checks, who decides and who is authorized to act. Examine quality, time, cost and the ability to detect and correct an error. The division of work will change when evidence justifies it. There is no need to defend a fixed list of things only a human can do. Responsibility for the consequences remains with those authorized to make the decision.

A study involving 758 consultants found that AI improved performance on tasks suited to its capabilities, but reduced the likelihood of a correct solution on a task chosen to fall outside its capability boundary. [3] The system and tasks examined do not define the boundary today. The finding explains why it must be tested in every consequential use.

Combining humans and machines also requires evidence. A review of 106 experiments found that, on average, the combination performed better than the human alone, but worse than the better of the human and the machine working separately. [4] The reviewed experiments were published between 2020 and mid 2023. We should add, cautiously and in parentheses, that AI capabilities now take a significant step forward every few months. When reading any study, we must therefore remember when it was conducted and which capabilities it examined...

This is a lesson about designing collaboration, not a ranking of current models.

Where appropriate, we will therefore compare several ways of carrying out the test. Include training, oversight and error correction in the cost. Seek evidence from reality beyond agreement among the answers of different systems. Those systems may draw on similar information and assumptions.

We will invest in stages that allow learning before expansion. At each stage, establish what we want to know, how much we are willing to invest, and what would justify continuing, changing or shelving. Examine the cost of capital and the effect on other possibilities too. A possibility can be good in itself, yet deserve resources less than another possibility available to the company.

Shelve the possibility without shelving the learning

Shelving is a decision to remove a possibility from active work and release the resources allocated to it. It must include a reason and accountability for implementing the decision. Stopping meetings alone sometimes leaves activity and costs continuing without a decision.

There are several distinct reasons for shelving. The possibility may not serve the foundational principle. An experiment may have disproved a condition essential to its realization. The expected contribution may not justify the resources and risk. A new capability in the market may have eliminated the advantage it rested on. We will record the precise reason, because it determines whether reexamination is worthwhile.

We will distinguish shelving from a planned suspension. A possibility may wait for a defined condition, such as an available capability or partner. For a suspension, we will establish the condition, who monitors it and when we will decide again. If we cannot define these, the suspension may conceal avoidance of a decision.

AI can help with shelving. We will ask it to present the strongest evidence against continuing, and examine whether our reasons rest on untested assumptions. We will also ask for the strongest argument in favor of continuing. A system asked only why we should stop may help us justify a decision we already wanted to make.

Human capital allows us to interpret the test and assess what remains valuable. Social capital allows us to say that a direction has failed without concealing evidence. Amy Edmondson’s study of 51 teams found a relationship between psychological safety and learning behavior. [5] I apply the lesson to the ability to propose stopping as well. The employee who exposed the failure needs to know that the organization wants to learn from it.

The person who proposed a possibility should not become the person who must defend it forever. We will require professional work, reliable reporting and compliance with agreed limits. We will also recognize the contribution of a justified experiment that led to timely shelving. When people are promoted only for possibilities that continued, an incentive to conceal reasons for stopping may arise.

However, trust does not remove the need for critical review. A cohesive team may defend its investment against an outside voice. We will include someone whose position does not depend on the possibility continuing, and, where needed, a customer or partner who does not accept its assumptions. We will assess continuation according to future benefits and additional costs. Money already spent does not, by itself, justify further expenditure.

Shelving also requires attention to the relationships created. We will explain the decision to employees and partners, and honor commitments or agree changes to them. A well managed stop can preserve willingness to work together in the future. Social capital can thus be preserved even when a particular possibility is closed.

Reexamine a possibility when reality changes

Advanced AI can change the cost of implementation, the time required or the ability to conduct a test. It can also enable competitors to provide the same contribution. Technological change may therefore justify reopening a shelved possibility, or shelving one that has already been advanced.

I suggest retaining a decision record for every shelved possibility. What did we want to make possible? What did we test? What did we find? Why did we shelve it, and what material change might justify another test? We will also retain knowledge, test cases and capabilities that can be used again. Shelving then ends the active investment while preserving a basis for future learning.

The possibility card from Chapter 3 now also becomes a decision record. We will record what changed and what we decided as a result. We will distinguish information available at the time of the decision from what became known only afterward.

An announcement of a new model is not enough to reopen everything. We will examine whether the change relates to the reason for shelving. If the bottleneck was the cost of analysis, a less expensive capability may be relevant. If the bottleneck was the absence of customer need, faster analysis does not prove that the need has emerged.

We will initiate another test with a limited budget and a defined evidence requirement. It does not automatically restore the possibility to a major investment program. Nor does retaining an idea justify ongoing attention to it. Organizational memory should be useful without becoming a permanent burden on the team.

The default I suggest is a brief weekly discussion of active experiments, a monthly review of relevant changes in capabilities, and a quarterly review of possibilities against the principle and available resources. The pace will be adapted to the activity and the cost of error. A material change requires a review even before the scheduled date.

We will attach a date and test conditions to every claim about a system’s capability. We will prefer to keep information and decision histories in a form that can be transferred between systems. The organization must be able to change tools without starting its thinking over again.

An example of creating, shelving and reopening

Consider an illustrative example of a small industrial component manufacturer. Its chosen purpose is to enable customers to operate equipment reliably. Its foundational principle requires adapting the solution to conditions of use, while preserving reliability that can be tested.

The team identifies a possibility to offer customized components in small quantities. An engineer brings design experience, and a service employee brings knowledge of failures. AI helps develop alternatives and test cases. A distributor connects the team to customers, and a laboratory supplies a complementary testing capability. The four components come together in creating and examining the possibility.

An experiment shows that demand exists, but design costs are too high. The company shelves the service in the form tested. It documents the cost and reliability requirements, and defines what cost reduction might justify reexamination. Employees move to another task, while partners receive an explanation and arrangements are made to settle commitments.

A capability later emerges that shortens the preparation of designs. The team tests it on the cases already examined. If costs fall without compromising reliability, another experiment with customers is justified. Price, product liability and required capital will also be examined. Reopening follows from a change in the bottleneck that led to shelving.

Suppose a further test shows that the cheaper approach compromises reliability. The company will shelve it too, because it does not meet the chosen principle. An impressive technological capability does not change the commitment to the customer.

Build an ongoing capacity to create and decide

People should develop through their work on possibilities. We will pair a subject expert with an employee experienced in working with AI. Learn to provide context, examine evidence and explain a decision. Preserve professional mentoring even when the system performs part of the work. Someone who does not understand the field will struggle to recognize when its proposal fails.

Employees need to see how sharing knowledge affects their role and future. We will build a path for their development and contribution to subsequent possibilities. If they believe that sharing knowledge will only be used to harm them, they will have a reason to keep it to themselves. Without an honest response to these interests, joint work will remain incomplete.

We will define rewards that recognize the creation of a possibility and the quality of its testing. Recognize knowledge transfer, the integration of a partner and early disclosure of a failure. Assess business success over the time needed for it to materialize. The number of ideas and the number of uses of AI are not sufficient measures.

To begin, I suggest a defined ninety day process. In the first month, we will formulate a principle and create several possibilities for realizing it. Choose a few for testing, according to their importance and the decisive assumption for realizing them. Allocate a team, budget and time, and define conditions for continuation and shelving in advance.

In the second month, we will conduct experiments with the people and partners required. Use AI to prepare alternatives and conduct critical review, and test the proposals against performance and responses in reality. Document what changed in our understanding, not only what was completed.

In the third month, we will decide what to continue, change or shelve. For each shelving decision, record the reason and reusable knowledge. If a suspension is justified, record a condition and a date for another decision. Actually transfer the released resources to the possibilities chosen.

The Board needs to see what the process created and what it released. Which possibility advanced, and what evidence justifies the next investment? Which possibility was shelved, and where were the resources transferred? What capabilities were built among people, and which partnerships enable realization? And what has changed in reality that requires a return to the inquiry into the principle?

Here, principles and practice come together. The foundational principle guides the company’s choice. Joint work expands possibilities and tests them. Decisions concentrate resources, and learning is preserved even after shelving.

Learning by Thinking Together

The future belongs not only to those who know how to solve problems, but to those who know how to create possibilities we have not thought of before. To realize them, we will also need to decide what to give up.

Even a stronger system does not provide knowledge of what will happen. It can broaden the search and assist with testing. Management must turn the combination of the system and people into an ongoing ability to create possibilities, choose among them and learn.

A question for the reader

If your foundational principle were already binding on you this week, which possibility would you start testing and which would you shelve, and whom would you bring into the thinking to make both decisions?

One Step into Practice

Choose an active possibility and honestly examine the justification for continuing. If you shelve it, actually transfer the resources to another task. Retain the reason for the decision and the condition that would justify reexamination.

Sources for further reading

[1] Dell’Acqua and colleagues The Cybernetic Teammate Organization Science 2026 The journal version describes an experiment involving 791 professionals. A study of integrating expertise and working with AI. https://doi.org/10.1287/orsc.2025.20702

[2] Anil Doshi and Oliver Hauser Generative AI enhances individual creativity but reduces the collective diversity of novel content Science Advances 2024 A story writing experiment on individual improvement and similarity among outputs. https://doi.org/10.1126/sciadv.adn5290

[3] Dell’Acqua and colleagues Navigating the Jagged Technological Frontier Organization Science 2026 An experiment on matching the use of AI to the task. https://doi.org/10.1287/orsc.2025.21838

[4] Michelle Vaccaro Abdullah Almaatouq and Thomas Malone When combinations of humans and AI are useful Nature Human Behaviour 2024 A systematic review and meta analysis of experiments in human and machine collaboration. https://doi.org/10.1038/s41562-024-02024-1

[5] Amy Edmondson Psychological Safety and Learning Behavior in Work Teams Administrative Science Quarterly 1999 A field study of psychological safety and team learning. https://doi.org/10.2307/2666999