This morning, I rented an e-bike for the semester.

The shop is a few minutes’ walk from my campus in Fuzhou. Within half an hour, I had chosen a bike, completed the formalities, and was making my way back to campus. In a city where e-bikes are part of the everyday traffic, the whole transaction felt routine.

What caught my attention was the owner, whom I know by his English name, Tom.

He looked understandably pleased with how things were going. When I first met him last year, his business occupied what could only be described as a hole in the wall. He had one assistant who seemed to perform every function in the company short of preparing an annual report.

Today, Tom operates from a smart showroom. He sells as well as rents e-bikes, has several warehouses, employs five people, and is now an authorized dealer for Ninebot. His English is good, which makes our conversations easier, but the vast majority of his customers are Chinese. The business does not depend on a small foreign-student niche.

Nor has every e-bike entrepreneur experienced that kind of growth. Tom strikes me as energetic, commercially astute, and alert to opportunities. His progress owes a great deal to his own judgment and effort. Yet personal qualities alone do not explain the opportunity that was there for him to seize.

What interested me, beyond the growth of one business, was what it revealed about how technologies become part of everyday life.

By 2025, China had about 380 million electric bicycles in use, roughly one for every four people. At that scale, the product is highly visible. What is less visible is the work that turns technical possibility into routine use.

We often use invention, innovation, adoption, diffusion, and scaling as if they mean the same thing. They do not. An invention creates a new technical possibility. Innovation puts a new or significantly improved product, process, or business model into practice. Adoption occurs when someone chooses to use it. Diffusion is the pattern created as those choices accumulate. Scaling expands capacity, output, or reach.

These processes are connected, but they are not interchangeable. An invention may never be widely adopted. Production may scale while demand disappoints. A useful product may diffuse slowly because repair, distribution, regulation, or user trust lag behind it.

The e-bike economy contains all of them. The vehicle reflects successive inventions and product innovations, not one dramatic breakthrough. Hundreds of millions of bikes in use is a measure of penetration, the outcome of decades of diffusion. The growth of production, distribution, and firms such as Tom’s is evidence of scaling. The surrounding system connects the two.

That is why the idea of an innovation ecosystem is useful, provided we use it precisely. A supply chain explains how a product is made and moved toward the customer. An ecosystem is broader. It includes the interdependent businesses, services, institutions, and infrastructure whose contributions must fit together before the customer can obtain the promised value.

Not every relevant factor belongs to the ecosystem, and not every factor is an innovation. Improved batteries, electronic controls, digital payments, connected features, and new rental or service models are innovations. Manufacturing depth, dealer coverage, and repair skills are capabilities within the ecosystem. Urban density and short travel distances are contextual conditions. Workable rules provide institutional support. What matters is how these factors interact.

A good e-bike is of limited use if it is too expensive, difficult to repair, inconvenient to charge, unsafe to store, or poorly matched to the journeys people make. Conversely, dense manufacturing and component networks lower cost and expand choice. Brands and distributors move products through national and local channels. Dealers hold stock, explain alternatives, arrange sales and rentals, and provide after-sales support. Repair shops and spare-parts networks keep vehicles on the road. Digital payments remove friction from small transactions. The surrounding urban environment gives people strong reasons to use the result.

Tom sits at the point where this system meets an individual customer. That does not mean we must label him an innovator. His significance is more concrete: he removes obstacles between availability and use. For someone staying in Fuzhou for a semester, buying may be unnecessary while a series of short rentals would be impractical. Tom turns that awkward middle case into a simple service. For other customers, he offers sales, recognized brands, maintenance, and a nearby person to return to when something goes wrong. He also learns from local demand and adjusts what he stocks and offers.

This is part of the work of diffusion.

An adoption decision occurs one rider at a time. Diffusion is the cumulative result when large numbers of people make similar decisions.

People make those choices when a technology seems useful, affordable, and safe enough compared with the alternatives. Its existence alone is not enough.

Scaling supports diffusion, but does not guarantee it. More factories, stock, shops, and service capacity make wider adoption possible; demand still depends on fit with people’s journeys, budgets, and routines. China’s e-bike story is powerful because supply and usefulness expanded together. The ecosystem helped connect capacity with demand.

Widespread use, however, exposes weaknesses as well as strengths. E-bikes bring battery-fire risks, illegal modification, road-safety problems, parking conflicts, and questions about battery disposal. A system can spread hazards as efficiently as benefits.

China’s revised compulsory safety standard, which took effect in September 2025, tightened fire-resistance, anti-tampering, and traceability requirements. This is institutional adaptation rather than innovation in the sense of a new battery or business model: the rules change in response to problems revealed by mass use. Without that corrective capacity, diffusion can undermine its own legitimacy.

Getting millions of people to use something is diffusion, not innovation. The innovation challenge is to develop or recombine products, processes, and business models that create practical value. The ecosystem challenge is to ensure that the complementary pieces needed to deliver it are present.

Managers should ask these questions separately. What is genuinely new? Whose behavior must change for it to be adopted? Which complementary products, services, capabilities, and rules are required? Which parts of the system must scale together? And what problems will only become visible after widespread use?

Tom’s business offers a small answer. Large systems depend on thousands of local acts of judgment: choosing stock, earning trust, solving mundane problems, hiring people, and noticing what customers need before a distant manufacturer can see it. He did not invent the e-bike. His contribution is to help the surrounding system work locally.

My transaction this morning felt ordinary because a great deal of coordination had already taken place out of sight. That ordinariness should not be mistaken for a definition of innovation. It tells us that diffusion is advanced and the supporting system has become dependable enough for use to feel routine.

The product is only the starting point. The harder managerial question is what must happen around it before people can simply get on and ride.

Dr. Nicos Rossides is a CEO, educator, and author whose work spans management, education, and artificial intelligence. He is Co-Founder and Chairman of the Advisory Board of listening247, a global insights firm, and Professor of Management at Minjiang University’s International Digital Economy College. His Routledge titles include Engaging the Workforce (2022), Employee Engagement in Startups (2025), and AI-Powered Insight (2026, with Michalis A. Michael).

The Geopolitical Economist is now at geopoliticaleconomist.com: analysis, research and a free weekly newsletter. Subscribe at geopoliticaleconomist.com/newsletter.

Writers: pitch us at geopoliticaleconomist.com/write-for-us.

The E-Bike Is Only the Visible Part was originally published in The Geopolitical Economist on Medium, where people are continuing the conversation by highlighting and responding to this story.

Most cooperation begins before every aspect of it has been fully articulated.

People agree on the broad purpose, establish responsibilities, and begin working together. Procedures develop as circumstances arise. Some are written down. Others take shape through meetings, emails, precedent, and experience. Over time, the partnership acquires a way of operating.

This is usually a strength rather than a deficiency. Few worthwhile collaborations could begin if every future situation had to be anticipated in advance. Flexibility allows people to adapt, build trust, and solve practical problems that formal agreements cannot prescribe.

But there comes a point when a process that has been clear enough to sustain cooperation is no longer clear enough for the demands placed upon it.

The problem is rarely that the parties follow entirely different processes. They agree on the broad sequence, the main responsibilities, and the intended outcomes. Yet important parts of the process remain underarticulated, and where the process is silent, people fill the gaps themselves. Without an articulated vision to refer to, several versions of the process can quietly coexist.

Cooperation then proceeds on maps that show much of the same terrain but mark its boundaries, routes, and landmarks differently. Most of the time this does not prevent progress. But when the partnership encounters an unusual case, a sensitive handoff, or a disputed responsibility, the differences suddenly become consequential.

Clear enough is not the same as shared

Processes rarely need to be perfectly specified before people can work with them. The broad direction is evident. Participants know which institutions and roles are involved, understand the main stages, and recognize the expected outcome. Experienced colleagues fill in the details through judgment, relationships, and accumulated knowledge.

This produces a reasonable sense that the process is adequately understood. But adequacy is a slippery standard. A process can appear adequately clear because routine cases continue to be handled, deadlines are generally met, and disagreements remain manageable. The absence of visible conflict is easily read as evidence of shared understanding.

Yet people may be relying on different assumptions about the aspects that matter most: who must be consulted before a decision is made; who has authority to approve, amend, or override; when responsibility passes from one party to another; which documents are controlling; what evidence is required; how exceptions are handled; and who resolves conflicting interpretations.

These differences stay hidden because routine activity does not test them. A familiar case follows the expected route. Someone notices a missing step. A colleague sends a reminder. A relationship bridges an institutional gap. The work proceeds, and the process appears more coherent than it is.

It is often the exception that reveals the map.

Consider a composite example. A manufacturer and a distributor have jointly served a market for years. A customer submits an unusually large warranty claim. The distributor’s service manager authorizes the repair, believing that authority over warranty work was delegated to her long ago. The manufacturer’s quality director expects to approve any claim above a threshold, and regards the earlier arrangement as an interim courtesy. The account manager who brokered the original understanding has since left. Each side can produce emails that support its reading. The claim is eventually settled, but the dispute costs more goodwill than the repair costs money. Nothing about the process had changed. It had simply never been shared.

Beyond tacit knowledge

The obvious explanation is that much of the process exists as tacit knowledge. People know from experience whom to contact, which matters require sensitivity, and how work actually moves across organizational boundaries. They understand the history behind arrangements that would be difficult to reconstruct from the written record.

This knowledge is valuable, and no process document can capture every judgment required in a complex organization. But tacit knowledge is only part of the issue. The harder problem arises when organizational knowledge is distributed unevenly, interpreted differently, or built on assumptions that have never been tested.

Three conditions deserve separate names. Some practices have never been articulated: the process depends on people remembering what to do and passing that knowledge on. Some knowledge is not equally shared: one group knows the formal requirements, another the administrative procedure, another the local operating convention, and no single participant sees how the parts fit together. And some understandings do not align: people are familiar with the same broad process yet interpret important aspects of it differently, each interpretation reasonable when viewed from within a particular role or institution.

A fourth condition compounds the others. An interpretation may simply be mistaken. It may have originated in an informal instruction, an old practice, or a partial reading of a requirement. Repetition then makes it feel established.

None of this requires bad faith. People act conscientiously on the information available to them. The difficulty is the absence of a shared mechanism for comparing interpretations before a problem forces the comparison.

The process is scattered across the organization

An organization’s operating logic is rarely held in one place. Formal agreements define the purpose and broad responsibilities of a partnership. Policies and regulations establish boundaries. Administrative documents describe particular procedures. Digital systems record transactions, approvals, deadlines, and outcomes. Emails preserve some decisions, while meetings produce understandings that never enter the formal record.

The process exists across all of these, and in the experience of the people who translate written arrangements into practice. They know which person should be informed even when formal approval is unnecessary, which step is required by regulation and which is merely customary, and how a broadly defined responsibility is divided in practice.

This human layer is not an unfortunate residue waiting to be eliminated by better documentation. It is part of how organizations function. The issue is one of proportion. When too much of the process depends on individual memory, informal interpretation, or personal relationships, cooperation becomes vulnerable. It works well while experienced people remain available and routine conditions prevail. It becomes unreliable when personnel change, workloads grow, or an unfamiliar situation arises.

What matters is how much tacit knowledge the process asks the partnership to carry.

Academic cooperation has its own operating logic

The challenge is particularly acute in academic institutions. Universities are not organized around the logic of standardized production. They depend on professional judgment, disciplinary autonomy, committees, and negotiated arrangements. A degree of ambiguity is even productive: it allows educators and administrators to respond to complex cases and exercise discretion.

But valuable discretion and avoidable ambiguity are not the same thing. Professional autonomy works best when people understand where their discretion begins and ends. Flexible procedures remain credible when participants know which requirements are non-negotiable. Without this clarity, flexibility gradually becomes dependence on local interpretation, and the process functions differently according to who is involved and which part of the institutional system they know best.

The aim of process mapping is not to make an academic partnership operate like a factory. It is to identify which aspects genuinely require judgment and which have remained ambiguous simply because they have never been articulated.

The added complexity of cross-border cooperation

In an international partnership, participants do not approach the process from a neutral starting point. They bring expectations shaped by their own institutions, administrative systems, and professional traditions, and they may use the same words while attaching different practical meanings to them.

Terms such as approval, supervision, coordination, responsibility, and compliance appear self-explanatory. Each contains considerable ambiguity. Does approval mean formal authorization, informed consent, or the absence of objection? Does supervision imply authority over the work, responsibility for its quality, or an advisory role? Does coordination mean arranging communication, controlling the process, or ensuring that others complete their assigned tasks?

Such differences should not be reduced to simplistic contrasts between national cultures. No country has a single institutional logic; universities, departments, and individuals vary greatly within any system. But international cooperation does bring distinct administrative histories into contact, and people naturally interpret an incompletely specified arrangement through the practices they know.

Legal and administrative frameworks add another layer. Some differences reflect preferred ways of working. Others arise from institutional policy, accreditation standards, national regulation, or academic requirements. A practice that seems needlessly formal may reflect a binding external requirement; a convention presented as mandatory may turn out to be local custom. Process clarification must therefore distinguish among matters the partners can agree between themselves, matters that require institutional authorization, and matters constrained by external legal, regulatory, or accreditation requirements. Without this distinction, the parties spend time negotiating what cannot be negotiated, or treating a negotiable practice as though it were fixed.

How apparent consensus forms

A false sense of consensus does not usually arise because people pretend to agree. It develops more subtly. The broad arrangement is accepted. Work begins. Participants interpret the gaps in ways that seem reasonable from their own positions, and because most cases can still be handled, there is little reason to compare notes. Familiar language deepens the effect: a meeting ends in apparent agreement because nobody has reason to unpack what each participant understands by consultation, review, or final approval.

Success masks the gaps. When conscientious people bridge unclear handoffs, chase missing information, and reconcile conflicting expectations, the process appears to work. Their competence reduces the immediate cost of ambiguity. It is a version of the blind spot I described in Looking Past the Visible: what functions smoothly is easily mistaken for what is understood.

Over time, practice acquires the authority of precedent. “This is how it has been done” begins to sound like evidence that this is how it was agreed. Yet precedent can reflect a sensible adaptation, a temporary workaround, or a misunderstanding that was never revisited. The fact that a practice has continued does not establish its status.

Two diagnostic questions cut through this. Where do people appear to agree only because the underlying assumptions have never been made explicit? And where does the cooperation depend on an individual’s interpretation rather than an agreed process? These are often the most consequential points in the entire map.

The right people, sufficiently engaged

There is a further complication. Discussions of informal coordination usually assume that experienced and capable people are compensating for gaps in the process. Sometimes they are. But weakly articulated processes can also place critical responsibilities in the hands of people who lack the relevant expertise, authority, or judgment, while those best placed to decide remain only loosely involved.

The risk is pronounced in academic institutions, where expertise is deep but narrow. Accomplishment in one discipline does not equip someone to manage academic processes or exercise authority outside it. What matters is the fit between the person, the task, and the decision being made, not anyone’s individual worth. A loosely specified process leaves that fit to circumstance: people act because they are available, because a role has been assigned to them, or because no one else has taken clear ownership, and over time their involvement acquires an authority that was never explicitly established.

Nor is capability alone enough. Those with the appropriate expertise and authority may still fail to give a matter sufficient attention, especially where senior or specialist roles span many responsibilities. Delegation is often necessary and effective. Without clear boundaries, however, critical judgments migrate to people who were never intended to make them.

Some process steps require routine execution; others require informed judgment; still others require the sustained involvement of someone with formal authority, because the decision affects standards, accountability, or institutional relationships. Treating the three as interchangeable creates risk. The consequences are rarely dramatic. They appear as inconsistent advice, duplicated work, delayed decisions, and recurring misunderstandings, and they accumulate, particularly when those with the relevant expertise engage only after a problem is visible. A sound process clarifies which decisions may be delegated, what competence delegated work requires, where final authority remains, and which issues need the direct attention of those ultimately accountable.

What process mapping is really for

I am currently leading an effort to map the joint operations of an international university partnership. The work has led me to see process mapping less as an exercise in drawing workflows and more as a method of organizational inquiry.

The first task is not to arrange activities into neat boxes. It is to establish what is actually known. What is formally required? What has been agreed? What happens in practice? Which elements are consistently understood, and which only appear to be? Where does practice rest on precedent rather than explicit agreement? Who is making critical decisions, on what basis, and are those people equipped, authorized, and sufficiently engaged to make them?

These questions show that the map cannot be extracted from existing documents or from one person’s account. It has to be developed through comparison. Different participants each hold a valid part of the process. The purpose is to assemble the accounts, identify where they converge, and examine where they do not, rather than to rule on whose is correct

Tone matters, because process ambiguity is rarely the fault of one individual or one side. It is the cumulative product of growth, staff changes, evolving requirements, and arrangements that developed incrementally. A useful map should not read as a catalog of deficiencies. It should create a common reference point from which the partners clarify responsibilities and strengthen cooperation. The exercise is best framed as a natural stage in the maturation of the partnership.

Three views of the process

The comparison works best when three views are kept deliberately separate.

The first is what is formally required: contractual commitments, institutional policies, legal and administrative obligations, academic regulations, and accreditation standards. These define important boundaries, though they rarely explain every operational step.

The second is what currently happens: formal activities alongside the informal coordination, reminders, and workarounds that have developed in practice. Current practice contains valuable knowledge. It can also contain duplication, gaps, and mistaken interpretations. Describing current practice is necessary. Endorsing it is not, and a map that merely reproduces current practice risks institutionalizing existing confusion.

The third is what should be agreed going forward: the future operating process, developed after the first two views have been compared. It should preserve useful flexibility while clarifying responsibilities, decision rights, information flows, handoffs, escalation routes, and the points at which senior or specialist attention is required. A formally correct process may be impractical; an effective current practice may lack proper authorization. The future process must draw intelligently from both.

Why handoffs deserve special attention

The most consequential gaps appear where work passes between people, units, or institutions. Within a team, shared context compensates for incomplete procedures. Across a boundary, assumptions are exposed.

A handoff is not merely the movement of a task from one box to another. It can involve the transfer of information, responsibility, authority, evidence, deadlines, and risk. Several questions need to be explicit. What must be passed on, and in what form? Who initiates the handoff? How does the recipient know that action is required? When is the sender’s responsibility complete? Who remains accountable for the final outcome? Who intervenes when information is incomplete or two instructions conflict?

Many process maps show handoffs as arrows without examining what the arrows contain. An undefined arrow may conceal the most fragile part of the process.

A map should reveal questions, not suppress them

Process mapping itself can create false confidence. A polished diagram makes an uncertain process look settled. Boxes and arrows impose order. Ambiguities disappear into short labels, and different interpretations are absorbed into language broad enough for everyone to accept. The result is visually clear while remaining operationally vague.

A useful map retains enough granularity to expose the questions that matter. For each important step, it should help clarify who performs it, who owns the outcome, who has authority to decide, what rule applies, where responsibility passes, and how exceptions are handled. The map does not need to answer every question immediately. But it should show where an answer is still required.

Where AI may contribute

AI can support parts of this exercise. Large language models can compare policies and guidelines, identify inconsistent terminology, extract references to responsibilities, and trace how a requirement is described across contracts, administrative guidance, and correspondence. This reduces some of the burden of analysis.

But AI cannot infer a legitimate shared process merely by combining the available text. The sources reflect different institutional perspectives, were written for different purposes, and often reproduce ambiguity rather than resolve it. Nor can AI decide whether a person is equipped to exercise a responsibility merely because their job title appears next to it. The work of interpretation, agreement, role design, and accountability remains human.

This is partly a problem of information architecture. AI performs poorly when policies contradict one another, terminology shifts across documents, responsibilities are described inconsistently, and the relationships between different sources remain unclear. It can help expose that disorder, but it also inherits it. A system that retrieves fragmented organizational knowledge more efficiently does not thereby turn it into a coherent operating process.

From workable cooperation to a shared understanding

A partnership can function well without every detail being formally prescribed, and its success may owe much to flexibility, goodwill, and the judgment of the people involved. Process mapping should not diminish those qualities. Its purpose is to ensure that cooperation does not depend on people carrying different versions of essential operating knowledge, or on critical decisions being made by those who happen to be available rather than those who are equipped and authorized.

A mature process does not eliminate discretion. It establishes where discretion is appropriate, who should exercise it, and when a matter requires a higher level of expertise or authority. The aim is a sufficiently shared operating understanding, supported by a credible alignment of roles, competence, authority, and attention, rather than perfect documentation.

The exercise is not an accusation. It is a recognition that cooperation evolves, institutions change, and arrangements that were once sufficient need to become explicit as a partnership grows. Process mapping, at its best, is how that maturation becomes visible.

Dr. Nicos Rossides is a CEO, educator, and author whose work spans management, education, and artificial intelligence. He is Co-Founder and Chairman of the Advisory Board of listening247, a global insights firm, and Professor of Management at Minjiang University’s International Digital Economy College (Fuzhou). His Routledge titles include Engaging the Workforce (2022), Employee Engagement in Startups (2025), and AI-Powered Insight (2026, with Michalis A. Michael).

Support writers. All our nonprofit’s offerings here.

The Process That Works Until It Doesn’t was originally published in Brain Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

A recent conversation with the photographer and educator Petros Nikolaidis left me wondering whether we have misunderstood what photographers actually do.

Like many conversations about artificial intelligence, this one began with technology. Generative AI can now produce remarkably convincing images in seconds: portraits, landscapes, editorial visuals, advertising concepts. The familiar question follows almost automatically. What happens to photographers when machines can create the images photographers once had to make?

Yet the conversation kept drifting away from images. For me it turned to the human skills and qualities that have seemed oddly absent from discussions of AI image generation: presence, patience, timing, judgment, context, and interpretation. Petros, a working photographer who also trained as a historian, came at the same subject from the other direction: photography has been through technological upheaval before.

The more we talked, the less the photograph looked like the whole of the work. That observation points to a question larger than photography. When technology reproduces an artifact, what happens to the practice that produced it?

Photography is usually described through its outputs, and the description has the shape of a tautology: photographers take photographs. True as far as it goes, which is not far. But this cliché is obviously incomplete.

A documentary photographer may spend days traveling to a location before taking a single image. A wedding photographer spends the day navigating relationships, expectations, and emotion. A portrait photographer establishes trust before ever pressing the shutter. Of all this, only the image survives. The observing, the waiting, and the judging leave little trace in the frame.

AI forces a distinction that has always existed but rarely needed stating: the difference between an output and a practice. An output is an artifact. A practice is the structured activity behind it, the preparation, attention, judgment, and interpretation that go into making it. Often the two overlap so closely that we stop noticing they are different things.

For most of the history of photography, that distinction hardly mattered. If you wanted a photograph, a photographer had to make one, and the photograph usually served as a proxy for the activity behind it. At first glance a photograph contains only what is visible: light, color, composition, subject matter. Yet we have long read more into it. We look at a war photograph and take it for granted that someone was there. We look at a portrait and assume some exchange took place between photographer and sitter. Until recently those were safe assumptions, because the image and the activity behind it were so difficult to separate.

Technology has often reproduced outputs before it reproduced practices. A convincing image can now be generated without traveling anywhere. A report can be produced without running the consulting project behind it. A set of lecture notes can be written without anyone having taught the class. The output can now appear without the practice, which means it no longer tells us as much as it used to about the activity that produced it.

Petros’s historical training is useful here, because photography has had this experience before. Roll film put the camera in the hands of amateurs. Retouching and darkroom manipulation raised the first disputes about what a photograph could be trusted to show. The move from film to digital displaced a generation of technical craft. Each time, a technology arguably separated part of the visible result from the skill behind it, and photographers had to work out what they were actually offering. How they described their own value at those moments, and what clients went on paying for, is a matter of record. Petros and I are looking at that record more closely, so I will not anticipate what it shows.

So the useful question is whether the picture exhausts the value of the activity that produced it. The answer will differ enormously by genre.

Consider product photography for an e-commerce catalog: five hundred items shot against the same white background, lit to show color, texture, and fit as accurately as possible. There is skill in that work, but the image comes close to exhausting the purpose of the activity. Nobody needs to have traveled anywhere, built trust, or interpreted a complex situation. The photograph is largely the point. It is therefore unsurprising that this segment may be particularly exposed to automation. Even here, though, the image has to answer to something outside itself. A catalog picture must show the particular item on sale, and a plausible object of the same kind will not do, so automating the picture still leaves someone to establish that it matches, and opens the door to errors.

Other kinds of photography are harder to call. Petros’s own answer to what the machine lacks is the one creative people often give: experience, or something close to soul. I am wary of the word, partly because it cannot be tested, but an example of his makes the point more precisely. One of his series connects photographs taken in Norway with themes from Greek mythology. A Norwegian landscape does not obviously suggest a Greek myth. The link comes from Petros, a Greek Cypriot who studied history in Greece, and it is the kind of link any of us might make when a place we are seeing meets an image or a story we carry from somewhere else. People do this so readily that we hardly notice. A machine can be asked to connect a fjord to a myth, and it may do so fluently. What it lacks, at least in the way a person has it, is a biography that makes one connection matter more than another, and so a reason to choose this link, this frame, and this sequence from the many available. That choosing is part of the practice, and in his series it seems to be where much of the value sits.

Richard Sennett made a version of this argument long before generative AI. In The Craftsman (Yale University Press, 2008), he argued that craftsmanship is more than the production of artifacts. It concerns the relationship between thinking and doing, between judgment and action. Seen this way, photography starts to resemble a practice as much as a product. Generative AI leaves that distinction intact and makes it harder to ignore.

Photography is only a particularly clear example. Consulting has the same structure. A client remembers the deck: the framework, the recommendations, the tidy chart. Yet much of the value may have resided elsewhere, in understanding which recommendation the organization could realistically absorb, and in building enough credibility that an uncomfortable truth could be said aloud. Neither appears on a slide, though both may have shaped every slide.

For decades, we often evaluated consulting through the artifact because the artifact was difficult to separate from the practice. A competent deck can now increasingly be produced from a competent brief. That does not prove the consulting activity was unnecessary. It forces us to ask where the value actually resided, a question I took up from the judgment side in Cheap Execution, Expensive Judgment.

Teaching has a similar structure. The slides are the artifact. The practice is reading a room of students, deciding what to leave out, and changing course halfway through a lecture because the room has told you something.

The larger consequence of generative AI may be that it weakens a relationship we had taken for granted. For a long time, artifacts and practices traveled together. The photograph implied a photographer at work, and the report implied a project behind it. That implication is becoming less reliable.

When technology reproduces the artifact, we are left asking a question that was previously easy to ignore: was the value located primarily in the output, or in the activity that produced it? Once an image can be generated without a camera, the debate is no longer only about images. It becomes a question of whether we treated the artifact as a proxy for the practice all along.

So what does a photograph contain? Light, color, and composition, certainly. Beyond that, it contains what we assume about the person who made it, and for a long time that assumption cost us nothing, because it was usually right. Now it has to be checked, and the answer will vary from one photograph to the next.

Dr. Nicos Rossides is a CEO, educator, and author whose work spans management, education, and artificial intelligence. He is Co-Founder and Chairman of the Advisory Board of listening247, a global insights firm, and Professor of Management at Minjiang University’s International Digital Economy College. His Routledge titles include Engaging the Workforce (2022), Employee Engagement in Startups (2025), and AI-Powered Insight (2026, with Michalis A. Michael).

A Message from AI Mind

Thanks for being a part of our community! Before you go:

• 👏 Clap for the story and follow the author 👉
• 📰 View more content in the AI Mind Publication
• 🧠 Improve your AI prompts effortlessly and FREE
• 🧰 Discover Intuitive AI Tools

What Does a Photograph Contain? was originally published in AI Mind on Medium, where people are continuing the conversation by highlighting and responding to this story.

We are taught early that understanding something means taking it apart. When a machine stops working, we inspect its components to find the fault. We often approach an organization in much the same way, examining individual performance or looking for a process that needs fixing.

Often, this works. A worn belt can explain an engine failure, just as a missing skill can explain why a task goes badly. Studying components and how they work together has helped us solve countless problems.

The difficulty begins when we assume that the explanation must always be found at that level. A group of accomplished musicians can give an uneven performance because they fail to coordinate. Recruiting a better violinist may leave the problem untouched. We need to understand what happens between the players as well as what each can do.

In such cases, learning more about each individual may get us only so far. We need to examine the interactions through which their efforts produce a shared result.

What taking things apart can tell us

Studying a system through its components is often described as reductionism. Its usefulness can encourage a stronger assumption: that enough detail about the components will provide every explanation we need. That assumption deserves scrutiny.

Of course, an engineer already looks beyond a parts catalog, tracing how forces travel through a structure and how one component affects another. The question is which interactions an explanation needs to include, and what we lose by leaving others out.

My friend and former colleague Fabio Montagnino and I occasionally spar over ideas, bringing his background in physics into conversation with mine in management scholarship. Reading an earlier draft of this essay, he pointed out that simple systems can also be highly organized. That observation sharpened my question: when does studying the parts separately obscure the behavior we want to explain?

A clock can be highly organized and still lend itself to being understood one mechanism at a time. What matters is how much the rest of the system affects the behavior of the part we are studying. If that behavior changes with its surroundings, studying the part in isolation may give us a misleading picture.

Physicist P. W. Anderson addressed this problem in his 1972 essay More Is Different. Knowing fundamental laws, he argued, does not mean we can reconstruct the behavior of complex systems from them. New levels of organization require their own concepts and explanations. Physics itself gives us reasons to question the sufficiency of explanations based on individual components.

The parts in their setting

In biology, a gene’s sequence is only part of what we need to know. Through gene regulation, cells control when a gene is expressed and in what amount. This helps them perform different functions and respond to changes in their environment. The sequence alone does not explain how that regulation works.

Studying a gene or protein in isolation can reveal an essential mechanism. We then need to establish how it operates within the living organism. Findings become misleading if we assume they apply unchanged under conditions we excluded from the study.

Consider two hospitals with similarly capable staff. For this thought experiment, assume that their resources and patient populations are also comparable. Yet one coordinates care reliably while the other repeatedly encounters delays.

Suppose that, in the second hospital, a test result reaches the ward after the team responsible for acting on it has finished its shift. Responsibility passes informally. The incoming team assumes someone else has followed up. Each person may be competent, yet the handover repeatedly fails.

Staff qualifications tell us little about the cause of the failure. Following the test result through the hospital would reveal where responsibility becomes unclear. Giving someone explicit responsibility for following up might improve care without changing the staff.

Calling this a problem of “culture” or “the system” would leave the explanation unfinished. We need to show how the arrangement contributes to the delay. Otherwise, we have merely exchanged one vague diagnosis for another.

A useful map leaves things out

The familiar observation that the map is not the territory helps here, provided we ask what a map is for.

A transport map leaves out most of the geography above ground. Its usefulness depends on preserving the connections a traveler needs. A complete list of stations would be far less helpful in planning a journey. Simplification can improve understanding when it preserves the features relevant to our question. A useful explanation, like a useful map, helps us see what would otherwise be difficult to follow.

An organizational chart, for example, may accurately show reporting lines while leaving out the colleagues people actually turn to for advice. To understand how a difficult decision gets made, we may need to follow those conversations. More detailed job descriptions would not tell us what we need to know.

When an explanation falls short, it is worth asking what kind of information is missing before collecting more of the same.

When measurement enters the loop

People can also respond to the measures used to assess their work, changing the very behavior we are trying to understand.

Imagine a customer-support team evaluated primarily on cases closed per day. Staff might begin prioritizing easy cases or closing difficult ones before the customer’s problem is resolved. The reported number improves. Some customers then return, creating additional work.

Managers see the new figures and adjust targets or staffing. Employees respond to those decisions, changing what the next set of figures will show. This creates a feedback loop in which measurement helps shape subsequent behavior.

Feedback can also help a team correct a problem. Colleagues might gradually develop a way of sharing difficult cases as they learn which approaches work. When patterns of coordination develop through such interactions without anyone designing the overall arrangement, we describe this as self-organization. A feedback loop alone does not establish that this has happened.

To understand a measure’s effects, we have to follow how people use it. The same formula might help one team identify problems while encouraging another to hide them, depending on how managers respond to the results.

Expectations that help shape events

A market forecast can influence behavior in a similar way. Suppose a respected analyst predicts a downturn. Investors who believe the forecast may sell, contributing to the decline they anticipate. Others may regard the fall as a buying opportunity and counteract it. The effect depends on how participants interpret and respond to the information.

This feedback between people’s understanding of a situation and the events their actions help produce is often called reflexivity. It complicates prediction because publishing a forecast may change the conditions on which the forecast was based.

Understanding this process still requires close analysis. We need to examine how expectations influence trading, and how price changes then affect expectations. Studying the transactions and following the wider feedback loop are complementary ways of investigating the same problem.

What organization explains

What counts as an adequate explanation depends partly on the question we are asking.

In the hospital, a staff member’s account might explain why they did not act on a particular test result. To explain why results repeatedly go unattended, we would need to examine the handover arrangements. Both explanations can be valid, but they answer different questions.

This raises a philosophical question. If changing the handover arrangements changes what happens to patients, should we treat organization itself as a cause? At the very least, the arrangements belong in our explanation because they shape what people know and who is expected to act.

Whether this explanation could ultimately be restated entirely in terms of individual actions, or even physical processes, is a further question. Our examples do not settle that debate. They show why understanding the arrangements matters to the problem we are trying to solve.

For managers, the immediate task is to investigate how the work gets done. A recurring failure may call for a change in how responsibilities pass between teams. A disappointing response to a new target may require us to reconsider what it encourages people to do. These possibilities become visible when we look beyond individual performance.

Studying the parts remains essential, but we need to return them to their setting to understand how they work together. The hospital’s unattended test result gives us a place to start. Following its journey may explain more than another assessment of the people who handled it.

Dr. Nicos Rossides is a CEO, educator, and author whose work spans management, education, and artificial intelligence. He is Co-Founder and Chairman of the Advisory Board of listening247, a global insights firm, and Professor of Management at Minjiang University’s International Digital Economy College. His Routledge titles include Engaging the Workforce (2022), Employee Engagement in Startups (2025), and AI-Powered Insight (2026, with Michalis A. Michael).

A Better Violinist Will Not Fix the Orchestra was originally published in Activated Thinker on Medium, where people are continuing the conversation by highlighting and responding to this story.

I had just finished a lecture on the attention economy when Teacher’s Day supplied an unexpected teaching aid. The lecture had treated attention in the customary way, as a scarce resource that platforms compete to capture and that analysts count in views, shares, and purchases.

My teaching assistant gave me a small figurine inspired by Niu Lai, the animated film usually rendered in English as Here Comes the Cow. She explained it first through a cheerful association: a bull market was coming. The name lends itself to that reading. In Mandarin the title echoes the phrase for a bull market arriving, and the Chinese financial press has reported investors adopting the film as a playful wish for rising prices. I had received a cultural reference and an optimistic financial outlook in one small object.

My students, who had sent warm wishes, filled me in on the film and its unexpected popularity. One of them was intrigued by how it had captured so much attention. I understood her question as being partly about its rough animation and apparently modest resources, though I would not put too much weight on my translation of her words.

Her question stayed with me. How does a film so much less polished than a major animated release become something everyone wants to discuss? And how does it then become a gift?

Reports of the film’s release help explain her surprise. The film opened in early August with no trailer and no marketing, and sold a few hundred tickets in its first ten days. Then clips of its simple computer models and awkward animation began to circulate. Early ridicule helped bring it to wider notice, and within weeks the box office had reached tens of millions of yuan.

Its budget is less clear. Published estimates vary, so I would hesitate to repeat any particular figure. More revealing is the director’s own account of the production: five years of work, hardware he could afford only secondhand, software written for architects rather than animators, and technical problems he had to solve himself. His mother, with no formal training, taught herself enough music to compose the score. The largest investment he describes is time.

Calling such a project “low cost” therefore needs some care. A modest cash outlay can sit alongside an enormous personal commitment. Five years does not become inexpensive because nobody submitted a timesheet.

Effort cannot settle the question of quality, of course. Teachers know this distinction well. “I worked very hard on it” deserves acknowledgment. Paragraph three may still need rewriting.

But knowing about the effort can change how we respond. An awkward scene might first seem funny because it falls short of what we expect in a cinema. Once we learn something about its making, we may also see the difficulty of attempting a feature film with so little. The scene has not changed. What we bring to it has.

That may explain some of the affection surrounding the film. Some viewers enjoy the absurdity. Others want to understand the jokes their friends are sharing. A single viewer can have several reasons to buy a ticket, and the popular story that “the internet mocked it, then fell in love” flattens all of this. The internet contains a great many people. Those who arrived to support the film were not necessarily the ones who first laughed at it. Mockery and affection can also coexist in the same person; anyone who has received a birthday card from a close friend will recognize the arrangement.

The film has also prompted discussion about AI. The director has said that no AI was used at any stage, and several commentators have taken the film’s roughness as evidence of that, a visible mark of human labor at a moment when polished images can be generated in seconds. I would not build too much on a single case. But it sits comfortably beside a point I have made elsewhere, that AI makes execution cheap while judgment stays expensive. As polish becomes easier to produce, it becomes a less reliable signal of skill or effort. Knowing how a work was made may then matter more to some viewers.

I had explored related territory in an earlier article, The Imperfect Appeal: Is China’s Labubu Heralding a Shift in Global Consumer Psychology?, where Labubu’s unconventional appearance offered a way to think about characters that depart from familiar ideas of beauty and cuteness. So the comparison came naturally. Here was another unusual creature that had become a shared reference, and an object whose significance might need explaining to anyone meeting it for the first time.

My students saw differences, and their response made me pause. “People like imperfect things” is a convenient explanation. It is also broad enough to hide quite different experiences.

Labubu’s oddness belongs to its design. Its teeth and expression are features of an imagined character, and an unconventional appearance can be executed with considerable skill. The film’s roughness is of another kind. Viewers are responding to limits in execution as well as to the character itself. Deliberately designing a strange creature and struggling to animate one are different creative situations. Both can produce affection, but we should not assume they do so for the same reasons.

The ways people encounter the two also differ. Collecting is central to Labubu’s appeal, with blind boxes, scarcity, resale prices, and the display of a new acquisition. Much of the cow’s appeal has developed through clips, reviews, jokes, and the story of its making. Both depend on conversation and commercial activity, but the emphasis differs. Calling both “imperfect” tells us little about those differences.

My TA’s gift brings the two together. The figurine on my desk is neatly made, yet it carries associations with the film’s roughness, the conversation around it, and the wish attached to its name. That helps answer my second question. My TA had found an object that let us share the joke and gave her good wishes a tangible form.

What does all this add to a lecture on the attention economy?

It reminds me that the things we count do not fully explain the experiences behind them. A view might register admiration or disbelief. A share might mean “this is wonderful,” “this is ridiculous,” or “you need to see this so we can discuss it.” A purchase might satisfy curiosity, support a creator, or make someone else smile.

A student of mine, asked to look into the film’s reception, began by asking two chatbots what people thought of it. The answers were fluent and plausible, but they were summaries of summaries. She had done what more of us now do, which is to send a machine to pay attention on our behalf, and it returned a tidy account with the disbelief, the affection, and the mixed motives pressed flat. To learn what audiences actually said, there was no substitute for reading what they wrote.

These distinctions matter because attention alone cannot tell us what people value. They also explain why technical polish is an incomplete guide to cultural success. People encounter a film through friends, clips, commentary, and stories about its makers, and they can enjoy the conversation as much as the work.

None of this makes roughness a dependable recipe. We hear about the unlikely production that becomes a sensation; many others remain obscure. Deliberately making something worse would be a poor lesson to take into either filmmaking or the classroom. What the cow offers instead is a reason to stay curious about what audiences find in a work, including things its makers never anticipated.

That curiosity seems fitting on Teacher’s Day. Teaching involves explaining, assessing, and occasionally persuading people that the reading is shorter than it looks. It should also include moments when a student’s question unsettles an explanation we thought was adequate. My student supplied such a question. The students’ wishes gave the occasion its warmth. My TA gave me something tangible to remember it by, along with a hopeful reading of its name.

Whether the craze lasts another month or grows into a lasting market for collectibles, I will remember the exchange with my TA and students whenever I see the figurine. The internet can move on. It will still be the cow my TA gave me, shortly after I thought I had finished explaining attention.

Dr. Nicos Rossides is a CEO, educator, and author whose work spans management, education, and artificial intelligence. He is Co-Founder and Chairman of the Advisory Board of listening247, a global insights firm, and Professor of Management at Minjiang University’s International Digital Economy College. His Routledge titles include Engaging the Workforce (2022), Employee Engagement in Startups (2025), and AI-Powered Insight (2026, with Michalis A. Michael).

Then My TA Gave Me a Cow was originally published in EduCreate on Medium, where people are continuing the conversation by highlighting and responding to this story.

My colleague Antonis and I sometimes discuss what would make humanity a more advanced civilization. A mathematician by training who teaches in Peru, Cyprus, and China, he often brings up the Kardashev scale, which ranks civilizations by the amount of energy they can harness. For him, the prospect of humanity one day commanding the energy resources of an entire planet, or even a star, shows how much more we might achieve.

I understand the appeal. With that much power, we could undertake scientific and engineering projects far beyond our present capabilities.

My reservation concerns what we count as progress. Would commanding a thousand times more power necessarily mean that people lived better lives? It would expand what we could attempt, but the achievement would depend on what we chose to do with it.

What Kardashev was asking

Kardashev was writing in the early space age, when radio astronomy was turning the search for extraterrestrial intelligence into an experimental science. In 1960, Frank Drake’s Project Ozma had begun listening for radio signals from nearby stars. The question was becoming practical: what could our telescopes detect?

In his 1964 paper, the Soviet astronomer Nikolai Kardashev examined how civilizations might communicate across interstellar distances. The power available to them mattered because it affected whether their signals would be strong enough to detect across those distances.

He distinguished three types of civilization. Type I used energy on a scale comparable to Earth at the time. Type II could harness the energy output of an entire star. Type III could do so on the scale of a galaxy.

The definitions have since evolved. Type I is now commonly described in terms of harnessing a planet’s collective energy resources. Carl Sagan introduced a continuous version of the scale in The Cosmic Connection (1973), using fractional scores rather than just three categories. He estimated Earth’s score at roughly 0.7, below the Type I threshold on the expanded scale. These later refinements explain why humanity is often described as having yet to reach Type I, even though Kardashev originally placed it there.

Kardashev also assumed that energy consumption would continue to grow far into the future. We can question that assumption while recognizing the practical value of his classification. Energy use can leave signals that astronomers might detect. It tells them much less about the lives of the beings using it.

Imagine astronomers on another world studying Earth. They might detect signs of industry and estimate our energy use. But could they tell whether that energy supported comfortable lives for most people or extraordinary privileges for a few? Even if they could, what would they regard as an achievement?

What energy use cannot tell us

Consider two imaginary societies with equal populations and the same total energy use. In one, reliable electricity supports comfortable homes, accessible education, and effective healthcare. In the other, much of the energy goes into coercion, wasteful production, and monuments to the people in charge.

The two societies could occupy the same place on the energy scale, although life in them would be very different, and it would be difficult to regard their achievements as equivalent.

To understand the difference, we need to ask four questions. How much energy is available? How efficiently is it used? What is it used for? And who benefits or bears the costs? These are questions of capacity, efficiency, purpose, and distribution. An energy ranking addresses capacity; it cannot settle the other three.

Efficiency matters, but it does not settle the other questions. A hospital and a surveillance system could both use power efficiently. We would still need to judge what they do and whom they serve.

This does not mean that lower consumption is always better. A community without reliable refrigeration, clean water, or adequate heating may need substantially more energy to live well. Asking it to celebrate a small footprint would confuse deprivation with restraint.

Additional energy can therefore have very different value in different settings. Reliable electricity could transform daily life in a village that lacks it; lighting empty buildings all night in an affluent city offers a much smaller benefit. In each case, we should ask what the extra energy achieves and whether the same benefit could be delivered with less.

AI and the limits of efficiency

Artificial intelligence shows why greater efficiency does not necessarily mean lower energy consumption. Better algorithms and hardware can reduce the electricity needed to complete a given task. But the total also depends on how many tasks we perform and how demanding they are.

When a task becomes cheaper, we may perform it more often or attempt work that was previously impractical. The savings on each task can be outweighed by the growth in use. William Stanley Jevons described a similar pattern in The Coal Question (1865): more efficient steam engines had helped increase Britain’s coal consumption because cheaper power found new uses.

The International Energy Agency’s 2026 analysis reports that energy use per AI task has been falling by at least an order of magnitude annually in recent years: a reduction of 90 percent or more. Yet electricity consumption at AI-focused data centers rose by 50 percent in 2025. Across all data centers, consumption rose by 17 percent. These figures describe different things: the energy needed for individual tasks and the electricity used by facilities handling a growing volume and changing mix of work. Together, they show why efficiency gains alone cannot tell us whether total demand will fall.

AI also depends on physical infrastructure. Models run on machines that need electricity and cooling, and manufacturing those machines requires materials and energy. More efficient software can reduce some of these demands; it cannot make them disappear.

Nor does rising demand today prove that it will grow indefinitely. Future consumption will depend on how widely AI is used, what people ask it to do, and the cost and availability of power and equipment. The value of that work is a separate question. Scientific analysis and disposable content can both consume electricity, but the meter cannot tell us what either contributes.

More than one possible future

The same uncertainty applies to the distant future. A civilization might need enormous amounts of energy for exploration or ambitious science. Another might stabilize its population and find that a smaller energy supply is sufficient for a high standard of living.

A 2024 study in the Astrophysical Journal explored the second possibility. Ravi Kopparapu of NASA’s Goddard Space Flight Center and colleagues modeled a solar-powered Earth with 30 billion people enjoying a high standard of living. Under the study’s assumptions, solar panels covering less than nine percent of the planet’s land area could meet their needs. Meeting those needs would not require stellar megastructures, the vast installations envisaged for collecting a star’s energy. Nor would their solar panels necessarily be easy to detect from another star system.

This is a possible scenario, not a prediction. Meeting everyday needs would not rule out exploration or other ambitions requiring much more energy. The study does, however, give us a reason to question the assumption that every prosperous civilization must follow the same path and keep increasing its consumption, or that we can judge its advancement from its energy budget. We cannot infer its future ambitions from its energy needs alone. A preference for restraint is no more a forecast than an enthusiasm for expansion.

How we judge progress today

The Kardashev scale offers an appealingly simple way to compare civilizations. More energy means a higher ranking. The temptation is to let that clear measure stand in for harder judgments about how well a society works for its people; if energy growth is the goal, then we can see a clear path forwards without having to first agree on its purpose.

We can see a similar temptation in the competition over artificial intelligence. Governments and companies announce new computing capacity, advanced chips, and the power plants needed to run them. These investments expand what can be done. They do not, by themselves, show how much public benefit will follow.

The IEA reports that delays in grid connections are prompting developers in the United States to build gas-fired generation on site for their data centers. This is a response to a practical constraint, but judging it requires more than counting the additional power. We also need to assess emissions, infrastructure costs, and the consequences for other electricity users.

The same applies to the computing capacity itself. Does it improve public services or produce useful knowledge? Who can afford to use it, and who pays for the infrastructure and environmental damage? Benefits and costs may fall on different people. A meaningful account of progress has to make those differences visible.

The next time Antonis brings up the scale, I will agree that commanding more energy expands what a civilization can do. Then I will ask how it uses that power and whose lives are better as a result. That is where our conversation about progress should begin.

Dr. Nicos Rossides is a CEO, educator, and author whose work spans management, education, and artificial intelligence. He is Co-Founder and Chairman of the Advisory Board of listening247, a global insights firm, and Professor of Management at Minjiang University’s International Digital Economy College. His Routledge titles include Engaging the Workforce (2022), Employee Engagement in Startups (2025), and AI-Powered Insight (2026, with Michalis A. Michael).

The Geopolitical Economist is now at geopoliticaleconomist.com: analysis, research and a free weekly newsletter. Subscribe at geopoliticaleconomist.com/newsletter.

Writers: pitch us at geopoliticaleconomist.com/write-for-us.

What the Kardashev Scale Leaves Out was originally published in The Geopolitical Economist on Medium, where people are continuing the conversation by highlighting and responding to this story.

A Chinese livestream host can appear to be doing something quite familiar: presenting products, answering questions, creating urgency, and encouraging viewers to buy.

It looks like selling. That description is not wrong, but it is incomplete.

The same host may also be testing customer reactions, identifying common objections, explaining unfamiliar features, shaping expectations, gathering product feedback, managing a community, and translating corporate language into something people are willing to hear.

The obvious question is no longer only whether the influencer is effective. It is where, exactly, the influencer belongs.

Marketing may hold the budget. Sales benefits from the conversion. Product teams learn from the feedback. Customer experience inherits the expectations. Public relations carries much of the reputational risk.

Everyone is involved, but no one quite owns the whole relationship.

The influencer does not quite fit the org chart.

The problem with calling it influencer marketing

The label sounds harmless enough. A company identifies a creator, agrees on content, measures reach and engagement, and hopes that attention turns into sales. Here, I use creator when emphasizing the person and the work; influencer when emphasizing the commercial role and its wider effects.

That is the familiar campaign model. The influencer sits outside the company and communicates something the organization has already decided. The product exists. The message exists. Then the campaign begins.

The sequence still describes many influencer relationships, but it describes China’s social-commerce environment less well.

On a livestream or social-commerce platform, discovery, explanation, demonstration, interaction, purchase, and feedback can occur in the same place. The creator moves between them without announcing that the job has changed.

One moment, the host is presenting the product. The next, she is answering a technical question, handling an objection, comparing alternatives, interpreting a complaint, signaling a price change, or telling the company what viewers do not understand.

The work has expanded. The organizational label has not.

Companies continue to place influencer activity under marketing because that is where it began. Yet the consequences now spread across product, sales, research, service, and reputation. The problem is no longer simply how to communicate through an influencer. It is who owns the work the influencer is doing.

Organizations manage the contract, not the role

Most companies know how to contract with an influencer. They can agree fees, deliverables, schedules, disclosure requirements, brand guidelines, and performance measures.

What they often manage less clearly is the role itself.

Is the influencer a media channel, a salesperson, a product demonstrator, a customer educator, a market researcher, a community leader, a brand representative, or an external adviser? In practice, the answer may be several of these at once.

That creates a familiar organizational difficulty. The relationship is assigned to one function, while its effects spread through several others.

Marketing may brief the creator without consulting customer service. Sales may celebrate the volume generated without asking whether the new customers are a good fit. Product teams may receive useful feedback informally, but have no process for deciding what to do with it. Service teams may face expectations created by content they never saw. Senior management may only become interested when something goes wrong.

The company has managed the transaction. It has not worked out how the relationship should operate across the business.

China makes the mismatch easier to see

China is not the only place where influencers perform several roles. Its social-commerce ecosystem simply makes the overlap unusually visible.

A creator may introduce a product, demonstrate it, answer questions, create scarcity, direct the purchase, and receive immediate audience reaction within the same commercial environment.

The customer journey is compressed. So is the organizational process behind it.

Activities that once occurred in sequence now happen together: market sensing, communication, persuasion, sales, customer education, feedback, and community management. The influencer stands in the middle of them.

This is why China serves as a useful backdrop rather than a claim about national exceptionalism. It reveals a wider change that is easier to miss where the stages remain more separate. China’s platform economy has a way of surfacing structural shifts early; I explored another instance in The Employee Who Never Sleeps, on the country’s virtual-employee economy.

China did not invent influence. Its platforms make it harder to keep pretending that influence belongs neatly inside marketing.

The creator as market sensor

One of the least appreciated functions of the influencer is listening. Companies usually imagine the influencer as someone who speaks to the market. Strong creators also hear things the organization may not.

Audience questions reveal confusion. Comments expose objections. Repeated requests point to unmet demand. Demonstrations show which features attract attention and which leave viewers cold.

A livestream can therefore produce an immediate stream of customer intelligence. It is not clean research. The comments may be noisy, emotional, incomplete, and unrepresentative. They are still evidence of what this audience notices, misunderstands, or wants.

The organizational weakness often lies in what happens next.

The creator reports comments to the marketing manager. Marketing passes selected observations to product. Product regards them as anecdotal. Customer service may already know the problem but remains outside the conversation.

Useful information travels through the company without acquiring an owner. The firm celebrates engagement while ignoring what the engagement is trying to say.

I made a parallel argument about employee listening in Surveys Spend Trust: once an organization asks, acting on what it hears stops being optional.

The creator as customer educator

Influencers also teach customers what a product is, why it matters, how it should be used, and what they should expect from it.

This matters most when the offer is unfamiliar, technical, culturally specific, or difficult to compare. The creator’s explanation may carry more weight than the company’s own because it uses the language, examples, and reference points of a particular community.

But explanation does more than improve understanding. It creates an expectation of the experience that will follow.

A product may be made to seem simpler, faster, more personal, or more effective than it really is. The influencer may not be dishonest. A smooth demonstration, a selected use case, and an enthusiastic audience can do the work without any explicit exaggeration.

The organization inherits the gap.

The campaign succeeds. The customer experience does not.

This is the point at which influencer marketing becomes an operational concern. The question is not only whether the message persuades, but whether the organization can deliver the experience the message has taught people to expect.

No number of likes can repair a promise the company was never equipped to keep.

The creator as salesperson

Livestream commerce also makes the sales role difficult to ignore.

The influencer creates interest, handles objections, frames the price, compares alternatives, establishes urgency, and closes the transaction. That looks remarkably like sales because, functionally, it is.

Yet the relationship is rarely managed as part of the sales system. The creator may not receive the same product training as internal salespeople. She may know little about fulfillment, returns, service, or stock availability. Her incentives may reward immediate conversion rather than customer fit.

A successful creator can therefore generate the wrong kind of success.

Sales rise quickly. Returns follow. Customer-service complaints increase. The company discovers that persuasion scaled more easily than delivery.

The front end earns the applause. The back end receives the bill.

The creator as brand interpreter

Brand managers often speak of consistency. Influencers trade in interpretation.

A creator does not merely repeat a brand message. She decides what deserves emphasis, what requires explanation, what can be ignored, and how the brand should fit into the audience’s world.

This is why good influencer content rarely sounds like a corporate statement. It is also why the relationship is difficult to control.

The company wants accuracy, consistency, and discipline. The audience values personality, candor, and independence. The creator is expected to provide both.

Too little control creates risk. Too much control produces content that sounds as though it has been approved by twelve people and enjoyed by none of them.

There is no tidy solution. The closer the creator moves to the organization, the more useful she may become. But the closer she moves, the harder it is to preserve the independence that made her persuasive.

Who owns the customer relationship?

The most uncomfortable question concerns ownership.

A company may generate substantial sales through an influencer while building very little direct relationship with the customer. The audience trusts the creator, follows the creator, and returns to the creator for advice. The brand benefits from that trust but does not necessarily possess it.

A company may believe it has built an audience when it has merely rented access to someone else’s community.

The difference becomes visible when the relationship ends. The influencer moves to a competitor and the audience follows. The creator launches a product of her own. A disagreement becomes public. A platform change weakens reach overnight.

What appeared to be a brand asset turns out to have belonged elsewhere.

China has already supplied the cautionary tale. When Dong Yuhui, the former English teacher whose scholarly selling style helped build East Buy into one of China’s most successful livestream operations, departed in 2024, he acquired the studio built around him and its established follower base. East Buy’s shares fell 23 percent the next day.

This does not make influencer partnerships unwise. It makes them more consequential than the campaign budget suggests. The stronger the creator, the more carefully the company must ask where customer loyalty actually sits.

The missing manager

The influencer affects several functions, but the company usually manages the relationship through one.

Marketing may coordinate the content, but it cannot alone decide whether product feedback should change the offer, whether customer expectations are realistic, whether operations can handle demand, whether service teams are prepared, whether sales incentives attract the right customers, or whether dependence on one creator has become dangerous.

These decisions cross departmental boundaries, precisely where organizations are often at their weakest.

Everyone has a view. No one has the full mandate.

The influencer becomes an external actor moving through an internal system that remains fragmented. That is the organizational problem hidden inside influencer marketing.

Start with the work, not the follower count

Companies often begin by asking which influencer they should use. They compare reach, audience profile, engagement, reputation, and cost.

Those questions matter, but they come too early.

The first question should be:

What organizational work are we asking this influencer to perform?

If the creator’s role is mainly awareness, the relationship may remain close to conventional promotion. If the creator is expected to educate customers, generate market feedback, handle objections, support a launch, or manage a community, the relationship requires much more.

The company must decide which functions need to be involved, what information the creator needs, how feedback will be captured, who acts on it, how much autonomy is appropriate, what promises the organization can support, and who is accountable when the relationship crosses departmental lines.

Follower count cannot answer those questions. Neither can engagement rate.

The wrong box on the chart

The problem is not that companies fail to understand influencers. Many understand them very well. The problem is that they often manage a new role through an old box.

The difficulty is not new. Students of organizations have long examined roles that straddle the firm’s boundary. What has changed is how much of the business now crosses that boundary through a single external actor.

The influencer contract sits in marketing. The customer consequences sit elsewhere.

China’s social-commerce economy makes that mismatch unusually visible because the creator’s role is broad and the customer journey compressed. But the underlying issue extends beyond China.

TikTok Shop, creator-led brands, expert communities, affiliate commerce, and platform-based selling are moving creators closer to the processes through which firms understand demand, explain value, and serve customers.

The form will differ from one market to another. The organizational question remains: where does the influencer belong?

The honest answer may be nowhere in particular. That is why the role must be designed rather than simply assigned.

The influencer does not fit the org chart.

The mistake is assuming that the organization around the role does not need to change.

Dr. Nicos Rossides is a CEO, educator, and author focused on management, innovation, and the future of work. He is Co-Founder and Chairman of the Advisory Board of listening247 and Professor of Management at Minjiang University’s International Digital Economy College. His books include Engaging the Workforce (Routledge, 2022), Employee Engagement in Startups (Routledge, 2025), and AI Powered Insight (Routledge, 2026), co-authored with Michalis A. Michael.

The Influencer Does Not Fit the Org Chart was originally published in Write A Catalyst on Medium, where people are continuing the conversation by highlighting and responding to this story.

There are moments when an AI system can appear remarkably intelligent.

It can analyze a difficult argument, write useful code, identify patterns across a large body of material, or make an unexpected connection between ideas.

Then, sometimes in the next exchange, it does something astonishingly foolish.

It misreads a simple instruction. It invents a source. It overlooks an obvious contradiction. It expresses confidence in a conclusion that falls apart under basic scrutiny.

The usual explanation is that the systems are still immature. As they improve, the foolish mistakes will diminish and their performance will become more consistent.

That may be true. But another possibility deserves attention. Perhaps the real problem lies not only in the systems, but in the way we interpret them.

We have learned to treat fluency, reasoning, understanding, judgment, and reliability as signs of a single underlying faculty. When these qualities appear together in a person, we call that person intelligent.

AI unsettles this familiar pattern. It can display some of the outward signs very strongly while other expected qualities remain weak, unstable, or unclear.

The signs are still there.

They simply no longer add up in the way we expect.

Sutton’s bitter lesson

Richard Sutton’s well-known essay, “The Bitter Lesson,” offers a useful starting point.

His argument was that the most important advances in AI came from general methods that could exploit increasing computation. Systems built around learning and search tended to outperform systems in which human experts tried to encode their own knowledge directly.

The lesson was bitter because it reduced the importance of carefully designed human expertise. Progress came from scale, learning, data, and computation, not by mimicking and encoding human-like intelligence.

Recent AI appears to follow this pattern.

Large models have acquired abilities that were not individually programmed into them. They can translate, summarize, write code, answer questions, and move between subjects with surprising ease.

Some of these abilities seem to appear once systems reach a certain scale.

The comparison with phase transitions in physics is tempting. Water warms gradually, but at a certain point it changes state. Something similar may happen in AI as it scales. Quantitative increases can produce what looks like a qualitative change.

Yet we do not fully understand these thresholds.

A new capability may genuinely appear. It may also have existed in weaker form before we noticed it. In some cases, the way we measure performance may make the change look more sudden than it really is.

The word “emergence” is used to describe what we observe. It does not always explain why it happened.

Still, the practical result is striking. These systems can now perform tasks that we once treated as evidence of intelligence.

The difficulty begins when we decide what that evidence means.

What does performance tell us?

Alan Turing famously moved away from the question of whether machines can think. He proposed a practical alternative. If a machine can behave in ways that appear intelligent, perhaps that is enough.

This was an important shift: it replaced a difficult question about inner experience with something we could observe.

Modern AI gives Turing’s position new force.

When a system explains a complex idea, responds to objections, adapts its language, and produces a fresh argument, it becomes hard to dismiss the performance as trivial.

Yet the same systems also expose the limits of behavior as evidence.

They can sound thoughtful without being dependable. They can produce convincing explanations without recognizing where those explanations are weak. They can follow the form of reasoning and still fail to notice when the reasoning has gone wrong.

Their performance encourages us to attribute intelligence; their inconsistency makes that attribution uncertain.

The issue is not simply whether an answer is true or false.

An answer may be factually wrong. It can also fail because it is incomplete, poorly framed, based on weak evidence, or inappropriate for the decision at hand.

A response can be technically accurate and still be misleading. It can answer the question asked while missing the more important question that should have been asked.

This is why the problem cannot be reduced to truth alone. The deeper issue concerns how well the answer relates to evidence, context, relevance, and judgment.

The old question of understanding

John Searle’s Chinese Room argument still matters here.

In simplified form, Searle argued that a system could manipulate symbols correctly, given the proper instructions, without understanding what those symbols meant.

The argument has been challenged many times. One important objection is that even if a single part of a system does not understand, the larger system may still do so.

Even so, Searle identified a problem that remains unresolved: producing an appropriate response does not by itself prove that understanding is present.

Generative AI makes this question harder because its responses are flexible, contextual, and often surprising. These systems do more than retrieve fixed answers. They generate, combine, infer, and adapt. That does not prove that they understand, but it makes the claim that they are simply manipulating symbols less satisfying than it once was.

The question remains open.

Hinton’s challenge

Geoffrey Hinton takes a different view.

He has argued that it may be too easy to dismiss neural networks as systems that simply match patterns. They learn distributed representations. These representations capture relationships among words, concepts, objects, and situations. They can then use those relationships across different tasks.

Perhaps this deserves to be called a form of understanding.

The word “merely” often does too much work in discussions of AI. Human beings also learn through patterns, prediction, association, and memory. Describing AI as statistical does not settle the matter.

The harder issue is what we mean by understanding.

Is it the ability to use concepts flexibly? Does it require direct contact with the physical world? Must it involve consciousness? Does it depend on having stable beliefs? Must a system recognize its own uncertainty?

There is no settled answer.

This takes us into both epistemology and ontology. How could we know that a system understands? And what kind of entity would it need to be for understanding to belong to it?

A short article cannot settle these questions, but AI has made them increasingly difficult to avoid.

The human template

Part of the confusion comes from the way the concept of intelligence developed. We formed it mainly by observing other human beings.

In human life, fluent language, reasoning, memory, practical competence, understanding, and judgment often appear together within the same person.

Not always, of course. People can be articulate and mistaken. Knowledgeable and unwise. Clever and reckless. Even so, the qualities usually appear within a recognizable human whole. We therefore came to treat them as expressions of one underlying faculty.

AI presents a different case.

It can show fluency without stable belief, produce reasoning-like answers without reliable self-correction, and move across many domains yet still fail on a simple variation. It can generate something novel without having any real intention or sound confident without having good grounds for that confidence.

This may explain why AI can appear astonishingly intelligent and astonishingly foolish in the same conversation. The apparent contradiction could lie partly in the categories we bring to the system. We assume that when one familiar sign appears, several others must be present as well. AI suggests that this inference is no longer safe.

There is a further difference in how these capacities are acquired. A person who performs a task well has usually learned from doing it, and carries that learning into the next attempt. Most current systems do not, at least for now. They draw on far more text than anyone could read in a lifetime, yet they retain little from one working session to the next. In their case, competence has been separated from the experience that would normally produce it.

AI can generate a flood of plausible output; verification and judgment remain selective.

What is really coming apart?

It would be too strong to say that intelligence itself is coming apart. That would assume that intelligence is a settled and unified thing. We do not know that.

What is coming under pressure is our inherited concept of intelligence.

The term bundles together several capacities: language, memory, reasoning, adaptation, understanding, judgment, self-correction, and the ability to deal responsibly with uncertainty, just to name a few. We often speak as though these capacities naturally belong together. AI raises the possibility that they can appear in very different combinations.

This does not prove that intelligence is innately fragmented in some philosophical sense. It tells us something more modest, that the label may be too broad to explain what we are observing.

Calling a system intelligent can hide more than it reveals.

Beyond true and false

The most serious concern is often described as reliability. But even that word needs care. We do not simply need answers that are factually correct.

We need answers that are sufficiently complete, relevant, appropriately framed, and open to correction.

Consider a business decision.

An AI system may produce an accurate market analysis but ignore a regulatory constraint. It may summarize the available evidence correctly but fail to notice that the evidence is weak, or answer the question it was given while leaving the real problem untouched. It may generate a plausible strategic recommendation without understanding the political, cultural, or organizational conditions required to implement it.

None of these failures can be reduced to a simple true-or-false test. The issue is whether the output is adequate for the purpose it is meant to serve. That requires more than factual accuracy.

It requires context, framing, judgment, and an awareness of what may be missing.

The problem of the corpus

Part of the difficulty lies in the material on which these systems are trained.

A language model learns from human-produced content. That includes a wildly diverse set of sources: scientific knowledge, historical records, journalism, opinion, advertising, and propaganda, including all the associated human prejudice, repetition, and error.

The volume is enormous. But volume does not resolve the problem.

Prominence or prevalence does not guarantee a source lacks distorting bias, or even that it is accurate. A false claim may be widely repeated. A correct claim may remain obscure. A respected consensus may later collapse. A marginal view may eventually prove right.

The model learns patterns in what human beings have written. It does not encounter reality in a direct and uncontested form.

It can compare sources, detect inconsistencies, weigh evidence, and produce a plausible synthesis. But where the underlying body of knowledge is incomplete or mistaken, the model may simply reproduce the problem in a more polished form.

It can distil the corpus, but it cannot guarantee that the corpus is sound.

At some point, real knowledge requires contact with the world through observation, prediction, experiment, measurement, intervention, or consequence. Even then, interpretation remains necessary; evidence does not explain itself.

Why this matters in organizations

These questions are not only philosophical. Organizations are already deploying AI in their workflows, using it to prepare reports, analyze customers, write software, and even to support decision-making and generate strategic options.

The gains are real.

Many forms of cognitive work can now be produced more quickly and at lower cost. But the greater volume of plausible output creates another burden.

Someone still has to decide whether the question was framed correctly, whether the evidence is relevant, whether the assumptions are defensible, and whether the answer fits the situation. Someone also has to notice what is missing. And someone must accept responsibility when the conclusion is wrong.

The main danger is not simply that AI will make mistakes.

The more troubling possibility is that it will produce outputs that look polished, complete, and ready for use even when they are incomplete, poorly framed, or unsuited to the context.

This is why judgment becomes more important as AI improves. The systems may become better at generating analysis while remaining uneven at recognizing the limits of that analysis. The bottleneck then shifts from production to evaluation.

A better set of questions

Discussion of AI often returns to a simple question:

Is it intelligent?

The question may now be too blunt. It assumes that intelligence is one thing and that a system either possesses it or does not.

A more useful approach is to ask which capacities are present, under what conditions they appear, how far they generalize, what kinds of error they produce, what they fail to notice, and how well they deal with uncertainty, context, and framing. These questions are less dramatic, but they are more revealing.

AI may not be moving toward human intelligence along a single line. It may be developing a different cognitive profile, one in which some capacities become very powerful while others remain fragile, intermittent, or unclear.

That does not prove that AI is intelligent, nor that it lacks understanding. It tells us that the word itself may no longer be doing enough analytical work.

The deeper lesson is not that intelligence has literally come apart. It is that the signs we once used to recognize it no longer lead reliably to the conclusions we once drew.

AI has made the label harder to use casually.

That may be one of its most important contributions to our understanding of intelligence.

Dr. Nicos Rossides is a CEO, educator, and author whose work spans management, education, and artificial intelligence. He is Co-Founder and Chairman of the Advisory Board of listening247, a global insights firm, and Professor of Management at Minjiang University’s International Digital Economy College. His Routledge titles include Engaging the Workforce (2022), Employee Engagement in Startups (2025), and AI-Powered Insight (2026, with Michalis A. Michael).

Support writers. All our nonprofit’s offerings here.

When the Signs of Intelligence Don’t Add Up was originally published in Brain Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

Sales at a consumer brand fall by eight percent.

Within minutes, its AI tools produce several explanations. The brand is losing relevance. Prices have become too high. Distribution has weakened. Product quality has slipped. The wider economy is slowing demand. A final synthesis argues that all five are interacting.

Each account is plausible. Each can be supported with data, charts, and a coherent narrative. None is obviously sufficient.

The organization is not short of explanations. It is short of a reasoned basis for choosing what to believe and what to do.

This is where much of the current discussion about AI and management remains incomplete. We ask whether AI can analyze more data, identify patterns, or improve decisions. Increasingly, it can. McKinsey, for example, describes AI as a potential researcher, interpreter, thought partner, simulator, and communicator in strategy work. That is a useful account of its expanding analytical role. It does not settle the harder question that follows: what should the organization treat as a sufficiently grounded understanding of the situation?

That question cannot be answered by the model alone. Nor should it be answered by a heroic manager who claims to see the whole picture. The understanding required for action is distributed across people, functions, records, tools, and experience. It is often incomplete, contested, and shaped by interests as well as evidence.

Management has long had to act without certainty. AI does not remove that condition. It changes the speed, volume, and apparent authority of the claims competing for attention.

The situation does not arrive already explained

Data are often discussed as though they were neutral facts waiting to be discovered. In practice, data are traces of a situation, shaped by decisions about what to measure, how to classify it, whose behavior to record, and what to ignore.

A fall in sales is observable. Its meaning is not. The same number may point toward weaker demand, poor availability, a pricing problem, a competitor move, or a temporary reporting effect. Before the team debates what the data mean, someone has already determined which data reached the meeting and how they were displayed.

The evidence placed before management is therefore not the situation itself. It is a selected and structured representation of it. This does not make the evidence useless or merely subjective. It means that analysis begins with choices and assumptions that should remain visible.

John Dewey described inquiry as the effort to transform an uncertain situation into one sufficiently settled to support intelligent action. His point was not that managers can wait until uncertainty disappears. Inquiry is a disciplined attempt to improve the quality of the situation by observing, questioning, and testing.

The order of operations matters. As I argued in Prediction First, Data Second, useful inquiry tends to begin with a disciplined prior, a working expectation that the evidence can then correct, rather than with the data alone.

AI enters this process after, and increasingly during, the selection and organization of evidence. It can reveal patterns that people missed. It can also make the limits of the available representation easier to forget.

AI expands interpretive capacity, not certainty

The most important change may not be that organizations have more data. They have lived with data abundance for years. It is that AI makes it dramatically easier to produce analyses, alternative hypotheses, and persuasive explanations from that data.

This can improve inquiry. A management team can ask the system to challenge its favored explanation, identify missing variables, compare the problem with similar cases, or show how different assumptions change the conclusion. AI can widen the set of possibilities considered before the organization settles too quickly on the first acceptable story.

But the same capacity has a less reassuring side. A fluent explanation can appear stronger than the evidence behind it. Research on AI-assisted judgment has found that explanations can increase reliance on both correct and incorrect answers. Other work distinguishes explanations that sound plausible from explanations that faithfully reflect how a model reached its conclusion.

The problem is not that AI always misleads. It is that coherence becomes cheaper to produce. A polished account may be insightful, derivative, weakly grounded, or simply wrong, while sounding much the same in each case.

More interpretations can protect an organization from blind spots. They can also create the illusion that producing more accounts is the same as understanding the situation.

AI may also narrow interpretation rather than expand it. Teams using similar models, prompts, and corporate documents may receive multiple answers that are variations of the same underlying frame. Apparent diversity of output is not necessarily diversity of thought.

From competing accounts to a working understanding

An organization cannot investigate forever. At some point it must allocate resources, change a price, delay a launch, revise a product, or decide not to act. The need for commitment gives management a responsibility that cannot be reduced to producing more analysis.

That responsibility is not to declare what is finally true. It is to organize a process through which competing accounts can be surfaced, challenged, and compared, and then to authorize a working understanding proportionate to the stakes.

Herbert Simon’s work on bounded rationality remains useful here. Managers operate with limited time, attention, and knowledge. They cannot examine every possibility or calculate every consequence. But this does not mean accepting the first answer that appears “good enough.” It means matching the depth of inquiry to the importance, reversibility, and urgency of the decision.

A small pricing experiment may justify quick action and rapid learning. Closing a factory, entering a politically unstable market, or making a decision that affects people’s livelihoods demands a much stronger evidential basis and a wider hearing of affected perspectives.

The threshold is therefore contextual. What must be understood before acting depends on the potential harm, the cost of delay, the availability of feedback, and the ease with which the decision can be reversed.

Understanding is distributed, contested, and shaped by power

No manager owns the organization’s understanding of a situation. Frontline employees may see customer reactions before the data capture them. Technical specialists may recognize weaknesses hidden inside a summary. Local teams may understand conditions that headquarters treats as noise. Customers, suppliers, and external researchers may hold knowledge unavailable inside the firm.

Edwin Hutchins showed that complex cognition can be distributed across people, tools, and representations rather than residing in one person’s head. This is a useful way to think about organizations. Their capacity to understand depends on how these different contributions are connected, not on the brilliance of a single decision-maker.

The contributions are not equal in visibility or authority. Rank, professional status, incentives, and existing investments shape which explanations receive attention. A numerical analysis may outrank frontline experience because it looks objective. A senior executive’s favored account may survive despite weak evidence because too much has already been invested in it. An AI synthesis may acquire authority because it is comprehensive and fluent, even when its assumptions remain hidden.

This is why the passage from interpretation to action is not purely technical. It is also institutional. Management must ask not only which account appears best supported, but how it acquired the standing to guide action and whose knowledge may have been excluded.

Alignment around an action is not the same as agreement about reality.

A team may decide to proceed while preserving a serious dissenting view. A board may authorize a pilot precisely because the evidence remains inconclusive. A business may coordinate around one course of action without pretending that uncertainty or disagreement has disappeared.

Some disagreements cannot be settled by more evidence

Not every management dispute is a disagreement about facts. Two executives may agree on the likely commercial outcome and disagree about whether the risk is acceptable. A restructuring decision can involve questions of fairness, loyalty, and obligation that no forecast can resolve. A pricing decision may depend on whether the organization values immediate margin or long-term trust.

AI can clarify the trade-offs. It can model scenarios, identify affected groups, and test the consistency of an argument. It cannot convert a conflict of values into a purely factual problem.

This distinction matters because organizations often hide value choices inside analytical language. A recommendation may be presented as “what the data say” when the data have only estimated consequences. The decision about which consequences matter remains a human and institutional judgment.

Before committing the organization, management should therefore separate at least three questions: What appears to be happening? What might follow from different actions? Which consequences is the organization prepared to accept?

Action is part of inquiry

Yet “before” should not be read as suggesting a clean sequence in which understanding is completed and then action begins. Organizations often learn by acting. They test, observe consequences, and revise their understanding.

Karl Weick emphasized that people partly make sense of situations through what they do and what their actions bring into view. Action does not merely follow interpretation. It changes the situation and generates new evidence.

This is especially important when uncertainty is high. A reversible experiment may produce better knowledge than another month of abstract debate. The responsible question is not always “Do we know enough for the final decision?” It may be “What limited action would help us learn without exposing the organization to unacceptable harm?”

A responsible commitment therefore names the current working view, the assumptions carrying it, the evidence that would weaken it, and the point at which the organization will review the result. It does not confuse decisiveness with permanence. The required discipline is not endless caution. It is the ability to commit while remaining open to correction.

The danger of institutionalized misunderstanding

The most serious risk is not that one AI analysis will be wrong. Organizations have long made mistakes. The deeper danger is that a weakly grounded account becomes embedded in targets, budgets, workflows, and incentives before it has been adequately challenged.

Once that happens, the organization begins producing evidence that appears to confirm its own interpretation. A sales team told that the problem is weak execution is measured more aggressively. A product team told that quality is not the issue receives no resources to investigate complaints. A pricing strategy built around assumed price sensitivity changes customer behavior in ways later treated as proof of the original assumption.

AI can accelerate this process by turning a tentative account into a polished plan, presentation, and operating model within hours. The speed is useful only if the organization has equally strong ways to preserve uncertainty, expose assumptions, and revisit the account after action begins.

In earlier pieces, I examined how cheap execution can make judgment more valuable and how managers can lose the cognitive practice needed to supervise automated work. The concern here is different. It is how a provisional explanation acquires institutional force and begins shaping what the organization can subsequently see.

What management must organize

Management does not need to monopolize interpretation. It needs to create a process by which the organization can move from distributed and contested claims to responsible commitment.

That process should bring relevant perspectives into view, connect claims to evidence, expose assumptions, protect serious dissent, and distinguish empirical disagreement from disagreement about values. It should also establish when further inquiry is justified and when delay has become its own costly decision.

One practical way to make distributed understanding more visible is to map how an issue is interpreted and acted upon across the organization: which activities shape it, what information enters at different points, which people and systems influence its interpretation, and where commitments are made or actions initiated. The value of such a map lies not only in the finished representation, but in the disagreements and omissions that emerge while it is being constructed. It can reveal where different parts of the organization are working from different accounts of the same situation.

A map, however, is not the situation itself. It simplifies, selects, and gives some relationships greater visibility than others. Its purpose should therefore be to make assumptions examinable, not to give a contested understanding the appearance of settled fact.

For consequential commitments, managers should be able to answer several plain questions:

• What is our current working account of the situation?

• Which evidence supports it, and how independent are the sources?

• What important evidence is missing or excluded?

• Which competing explanation has not been adequately ruled out?

• What values and interests influence the choice?

• What would cause us to revise our view?

• Can we learn through a limited and reversible action before making a larger commitment?

These questions do not guarantee a correct decision. No process can. They make it harder for fluency, hierarchy, or institutional convenience to pass unnoticed as understanding.

Before the organization acts

AI will continue to expand the analytical reach of organizations. It will generate hypotheses, compare scenarios, and challenge established views. In many cases, it will improve the quality and speed of inquiry. But improvement is not automatic. A recent meta-analysis found that combinations of humans and AI often perform worse than the best of either working alone, and that the gains vary with the task. Whether an organization captures them depends on how the collaboration is organized.

The advantage will not come from having the most analyses or the most eloquent explanations. It will come from building organizations capable of deciding what those analyses warrant, what remains unknown, and what level of commitment the available understanding can responsibly support.

Management does not own the truth. It organizes the passage from contested interpretation to organizational commitment, and keeps that commitment open to correction.

Before the organization acts, it must decide what it understands well enough to put its name, resources, and future behind.

Related writing

This argument develops a different part of a wider discussion I explored in Cheap Execution, Expensive Judgment, The Quiet Deskilling of Leadership, When Confidence Outruns Competence, When Algorithms Sit at the Table and Roots, Trunks, and the Seduction of the Leaf. Those pieces focus on judgment, capability, and the development of expertise. The present article focuses on the organizational passage from competing claims to action.

Dr. Nicos Rossides is a CEO, educator, and author focused on management, innovation, and the future of work. He is Co-Founder and Chairman of the Advisory Board of listening247 and Professor of Management at Minjiang University’s International Digital Economy College. His books include Engaging the Workforce (Routledge, 2022), Employee Engagement in Startups (Routledge, 2025), and AI Powered Insight (Routledge, 2026), co-authored with Michalis A. Michael.

A Message from AI Mind

Thanks for being a part of our community! Before you go:

• 👏 Clap for the story and follow the author 👉
• 📰 View more content in the AI Mind Publication
• 🧠 Improve your AI prompts effortlessly and FREE
• 🧰 Discover Intuitive AI Tools

Before the Organization Acts was originally published in AI Mind on Medium, where people are continuing the conversation by highlighting and responding to this story.

In an earlier article, I used the metaphor of the pilot and the detective to describe two different ways of navigating judgment under pressure, especially in knowledge work.

The pilot acts before everything is known. He reads the instruments, accepts uncertainty, makes a call, and adjusts course as new information comes in. The detective resists premature closure. She notices the clue that does not fit, reopens the file, worries about the missing fact, and refuses to sign off until the matter has been properly understood.

Both figures are necessary. A world made only of pilots would be reckless. A world made only of detectives would probably never leave the ground.

The difficulty begins when we put detectives in systems designed for pilots and then call them slow.

This is not simply a story about AI, although AI now makes the tension more visible. Long before generative AI entered the workplace, some people already worked at a different internal resolution from others. They did not merely want to finish the task. They wanted to know that it had been done right. They did not skim for obvious errors and move on. They checked, cross-checked, worried, returned to the evidence, and often found things that others had missed.

In many modern workplaces, this kind of mind is both needed and punished.

It is needed because organizations say they care about accuracy, compliance, quality, responsibility, and truth. It is punished because many knowledge-work systems are increasingly built around volume, throughput, dashboards, responsiveness, and speed. The result is a strange inversion. The person with the highest standards may appear to be the least productive.

That is the integrity tax.

The tax is paid by people whose internal verification threshold is higher than the system’s. Others aim for plausible completion. They ask, “Is this good enough to move forward?” The high-resolution thinker asks a different question: “Can I be sure?”

That difference can be costly.

To the system, the cautious worker may look inefficient. To the worker, the system may look morally careless. Each misunderstands the other. The organization sees hesitation where there is conscientiousness. The individual sees compromise where the organization sees necessary triage.

This is not a small personality difference. It is a difference in calibration.

Some people are comfortable operating in grey areas. They know that much of professional life consists of approximations, prioritization, and imperfect decisions. They are not necessarily careless. They have simply learned that work often requires action before certainty.

Others experience the grey area differently. For them, uncertainty is not merely a condition of work, but a warning light. They cannot easily reduce the resolution of their attention, or pretend not to see the ambiguity once they have noticed it. Their integrity is tied not only to the final answer but to the process by which that answer was reached.

The 8K mind does not easily become 1080p simply because the workplace finds 8K inconvenient.

This is not equally true of all work. Some institutions still preserve slower rhythms. Large public bodies, universities, regulators, international agencies, and certain professional bureaucracies may offer pockets where careful procedure is valued and speed is not the only currency. Although even there, the pressure to respond faster has grown. Emails multiply, documents circulate, deadlines tighten, and technology accelerates expectations.

But the mismatch becomes especially acute in sectors where knowledge work has been reorganized around volume: consulting, publishing, media, law, finance, technology, compliance, research support, and many forms of professional services. These are domains where judgment is demanded, but speed is often rewarded more visibly than judgment itself.

The exasperated question is easy to imagine: how is one supposed to keep up without becoming less serious about the work?

AI sharpens the question.

For the pilot, AI can feel like an escape hatch. It accelerates drafting, summarizing, comparison, synthesis, and review. It allows the user to move more quickly through tasks that once consumed too much time. The pilot understands its limits, at least in principle, but is willing to treat it as a co-pilot: useful, fallible, and worth checking.

For the pure detective, AI may feel different. While a useful tool, it is also another source of uncertainty. Where did the answer come from? What has been omitted? What has been smoothed over? What false confidence has been introduced? The pilot sees acceleration. The detective sees possible contamination.

The issue, then, is not speed itself. It is the quiet substitution of speed for quality. Faster search, faster drafting, faster comparison, and faster synthesis may be valuable when they improve judgment. But faster thinking is not necessarily better thinking. The detective’s hesitation is partly a refusal to let acceleration masquerade as understanding.

This difference is easy to caricature. The AI enthusiast may see the manual knowledge worker as stubborn, sentimental, or afraid of modern tools. In return, the detective may see the AI enthusiast as shallow, opportunistic, or too willing to outsource judgment. Both views miss the point.

The deeper issue is that people vary in how they relate to uncertainty, and organizations rarely design work with this variation in mind.

The high-resolution mind is not always suited to high-volume work. In fact, putting such a person into a production-line integrity role can be quietly destructive. The work may appear to require care, but the real reward system values speed. The person is asked to protect standards while being measured by throughput. They are told that quality matters, but the spreadsheet says otherwise.

Over time, this can produce a particular kind of burnout. It is not caused only by workload, but by moral friction. The worker feels forced to choose between doing the work properly and doing enough of it to be seen as competent.

The obvious managerial response is to say: learn to prioritize. And often that advice is valid. Professional life does require judgment about where to spend attention. Not every task deserves forensic scrutiny. Not every email is a legal affidavit, and not every ambiguity is a crisis.

But there is another possibility we discuss less often. Perhaps some people should not be trained to become faster detectives. Instead, perhaps they should be moved into rooms where detective work is actually valued.

There are domains where slowness is not a defect. Bioethics, forensic law, high-stakes compliance, safety engineering, deep-tech verification, medical review, financial audit, academic peer review, and certain forms of investigative work all depend on people who are bothered by inconsistencies that others ignore. In such settings, “good enough” can be dangerous. Skimming is not efficient; it is negligent.

The problem is that many organizations fail to distinguish between two kinds of integrity work. One is high-speed triage, where the task is to process many cases and escalate the few that look dangerous. The other is high-stakes judgment, where the cost of error is severe and the point is not to move quickly but to be right.

The same person will not necessarily thrive in both.

We talk a great deal about cognitive diversity, but often in a vague and comfortable way. We like the phrase when it suggests creativity, personality balance, or different perspectives around a meeting table. We are less comfortable when cognitive diversity creates operational inconvenience. The high-resolution thinker is admired in principle but marked down in practice.

This is especially true in systems built around metrics. Metrics flatten differences. They make work visible, but not always intelligible. A dashboard can tell us how many cases someone processed, but it may not tell us how many errors they prevented, how many hidden risks they noticed, or how much quiet responsibility they carried.

The 8K mind does not look impressive on a 1080p screen.

There is also an educational dimension here. Some elite forms of higher education reward close reading, argument, evidence, and precision. Students are trained to notice weak claims, unsupported leaps, and hidden assumptions. They learn that serious work requires attention. Then they enter workplaces that often reward speed, responsiveness, and a willingness to move on.

The very habits that once signaled excellence may now create difficulty.

This does not mean that universities are wrong to train rigor. Nor does it mean that workplaces are wrong to value speed. The difficulty lies in the mismatch. A person trained to respect complexity may struggle in a role that treats complexity as a processing obstacle. A person whose mind is built for depth may suffer in a system that keeps asking for breadth at speed.

There is a temptation to describe such people as belonging to a previous era. That is too easy, and probably wrong. Earlier eras were not necessarily kinder to reflective people. They had their own hierarchies, exclusions, and forms of pressure. What may be different now is the combination of speed, visibility, comparison, and measurement. The reflective worker is not merely asked to think faster. She is constantly shown that others appear to be producing more.

AI may intensify this divide.

For some workers, it will help bridge the gap between human capacity and organizational demand. It will allow them to draft faster, search more widely, compare more quickly, and check more efficiently. For others, it may increase anxiety because every AI-generated output becomes another object requiring verification. The tool that liberates one person may burden another.

This is why the future of work cannot be understood only in terms of skills. We also need to understand temperament, standards, and moral psychology. The question is not simply whether people can use AI. It is whether they can use it without feeling that they have betrayed the kind of worker, or person, they are trying to be.

Managers need to become better at spotting this distinction.

When someone is slow, the first question should not be whether they lack ability. It should be: what kind of slowness is this? Is it confusion, avoidance, poor training, perfectionism, conscientiousness, fear, or genuine depth of perception? These are not the same thing. Treating them as identical leads to bad management.

A high-precision instrument makes a poor hammer. That does not make it useless. It means one has misunderstood the instrument.

The practical challenge is to design better fits between minds and roles. Some people need coaching to reduce unnecessary checking. Some need clearer thresholds for when enough is enough. Some need tools that help them verify faster without feeling that they are surrendering judgment. Some need to leave volume-based work altogether and move toward lower-volume, higher-stakes environments.

The worst solution is to shame the detective for not being a pilot.

Organizations will continue to need pilots. They will need people who can act under uncertainty, absorb new tools, improvise, and keep things moving. But they will also need detectives, perhaps more than ever. In a world of synthetic text, automated decisions, AI-generated summaries, and confident machine error, the people who pause may become inconveniently valuable.

The danger is that we exhaust them before we learn where to place them.

The future of knowledge work will not belong only to the fast. It will belong to organizations that know when speed is the point and when speed is the problem. That requires a more subtle understanding of human capability than productivity metrics usually allow.

The reflective mind is not obsolete, but it is badly placed when systems confuse pace with value.

The integrity tax is not paid because someone cares too much. It is paid because the system does not know how to price care properly.

And that may be one of the great management failures of our time.

Dr. Nicos Rossides is a CEO, educator, and author focused on management, innovation, and the future of work. He is Co-Founder and Chairman of the Advisory Board of listening247 and Professor of Management at Minjiang University’s International Digital Economy College. His books include Engaging the Workforce (Routledge, 2022), Employee Engagement in Startups (Routledge, 2025), and AI Powered Insight (Routledge, 2026), co-authored with Michalis A. Michael.

Support writers. All our nonprofit’s offerings here.

The 8K Mind in a 1080p Workplace was originally published in Brain Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.