A Note for Global AI Cinema
I am especially pleased for this essay to appear in the context of Global AI Cinema, whose nonprofit mission I strongly support. Global AI Cinema begins with a question very close to the one that motivated this essay: if AI can generate millions of films while human attention cannot scale, who will define excellence, meaning, and cultural value? Its commitment to combining scalable technical assessment with human artistic judgment addresses one of the central problems of the AI cinema era: abundance makes trustworthy evaluation more, not less, important.
I see that mission and the argument of this essay as complementary. A trusted public evaluation system asks which films are excellent and which works deserve recognition, preservation, and cultural significance. This essay asks a second question that emerges immediately afterward: even among excellent films, which work is right for this particular person, at this particular moment, and which future works should people like this person help bring into existence?
Public evaluation and personal matching are not substitutes. We will need both. Institutions such as Global AI Cinema can help humanity establish shared standards of excellence and meaning, while Personal Agents and Agentic Media may help individuals navigate that increasingly abundant cultural world according to their own time, interests, values, privacy, and agency. The first helps society discern value; the second may help each person act on value. Together, they point toward a much larger possibility: AI may not merely change how movies are made. It may force us to invent an entirely new media architecture.
1. What If 10,000 Movies Were Born Every Day?
I have been thinking about a simple question: what happens if AI makes filmmaking dramatically easier? Traditionally, making a movie requires writers, directors, actors, cinematographers, lighting crews, sets, post-production, capital, coordination, and time. Today, AI filmmaking tools are already moving from text and reference images to shots, and from shots toward scenes. There is still a large gap between generating clips and making a truly excellent film, but the technical threshold for content production is clearly changing.
The obvious conclusion is optimistic. Production costs fall, more people gain access to filmmaking, and more films, micro-dramas, and videos are created. Historical epics, fantasy, science fiction, and other genres that once required expensive sets, costumes, locations, and visual effects may become accessible to much smaller teams. People who previously had no realistic path into the film industry may gain a new way to express themselves. That is genuinely exciting.
But let us push the trend further. What if we do not get ten more movies a day, but 10,000? The number is not meant as a forecast of current output. It is a thought experiment designed to ask what happens to the media system when production capacity is no longer the primary constraint.
A human being still has only 24 hours in a day. Even if I spent only one minute watching a trailer for each of 10,000 movies, the task would be impossible. We can give machines more compute, but we cannot give every viewer ten additional evenings. Making a work easier to produce does not make it easier to discover, and being discoverable does not mean it will actually be watched.
So the deeper question for AI Movie may not be how to make the movie. It may be how to reorganize the relationship between content and people when the supply of content vastly exceeds human capacity to consume it. To answer that, we need to look beyond generative models and return to media history.
2. Five Hundred Years Ago, People Were Already Complaining About “Too Many Books”
Information overload was not invented by the internet. As printing expanded across Europe, the scale of copying and circulating books changed dramatically. We usually tell this story as one of intellectual progress, and rightly so. But from the reader’s point of view, it also introduced a new and less comfortable experience: more books were available than any individual could read, remember, or judge.
Historian Ann Blair explored this phenomenon in Too Much to Know. European scholars in the sixteenth and seventeenth centuries were already complaining about an excess of books and developing new methods for excerpting, organizing, classifying, and retrieving knowledge. The experience of information overload is older than modern media and is not unique to one civilization.
The important lesson is not that people in the past were anxious too. The lesson is that reducing the cost of producing and accessing information does not automatically solve the problem of using information. Once books became easier to print, the next difficulty was how to find the right book, judge it, and place it within an existing body of knowledge. Catalogues, indexes, summaries, bibliographies, reviews, libraries, and publishing reputations all became more important. None of these looked as revolutionary as the printing press itself, yet they played a major role in determining whether printed knowledge actually entered human life.
This suggests a pattern that runs through media history: technology often does not eliminate scarcity; it moves the dominant scarcity somewhere else. At one point, the difficult thing was obtaining a book. Later, the difficult thing became choosing which book was worth reading. Success on the production side gradually created a new problem on the receiving side.
3. Why Scarcity Keeps Moving Toward the Human
The later history of media can be understood in much the same way. Printing expanded copying. Radio and television expanded broadcast reach. The internet reduced the cost of cross-border distribution and connection. Social media lowered the threshold for ordinary people to publish continuously. A person no longer needed a newspaper, a television station, or a publishing house to send writing, audio, or video into public space.
User-generated content brought an extraordinary democratization of media, but it also created the other side of the equation: content abundance. High-quality content did not disappear. In fact, there may be more excellent content today than at any time in history. The problem is that it exists inside a vastly larger sea of ordinary, repetitive, sensational, derivative, or low-quality material.
The center of value in the media industry therefore began to move. At one stage, the key problem was production; then it became distribution; later, increasingly, it became discovery and recommendation. One of the fundamental problems solved by YouTube, TikTok, and other content platforms is no longer whether content exists. It is what to show a user next among nearly infinite content.
Herbert Simon identified the underlying paradox decades ago: information consumes the attention of its recipients, so abundance of information creates scarcity of attention. He also warned that we should not count only the cost of producing and transmitting information; we must also count the human time required to receive it.
Generative AI pushes this imbalance further. Social media allowed almost anyone to create, but writing, shooting, editing, and producing still demanded sustained human labor. Generative AI begins to reduce the cost of many of those production steps and makes supply less constrained by the creator’s own time. That does not mean generation has literally become free. Regenerating a video is not the same as copying a finished video. Compute, energy, creative judgment, and human direction still have costs.
But the structural problem appears long before those costs reach zero. It is enough that content supply grows much faster than human viewing capacity. In the past, we worried that people did not have enough content to consume. In the future, we may have to confront the opposite problem: even when good content is abundant, people may still be unable to find what is truly worth their time.
4. Public Evaluation and Personal Matching Solve Different Problems
When content becomes excessive, the obvious response is to improve evaluation and identify the best work. Film awards, critics, review sites, ratings, festivals, and organizations such as Global AI Cinema all play an important role here. They help establish public standards, recognize creative achievement, and protect meaningful work from being overwhelmed by sheer volume.
In the AI era, that public evaluation function becomes even more important. If millions of works can be generated, society needs institutions capable of asking which films demonstrate genuine artistic achievement, originality, cultural significance, and human impact. Global AI Cinema explicitly places this human judgment at the center of its mission.
But another problem remains. There is a crucial distinction between saying, “This is a good movie,” and saying, “This is the right movie for me to watch right now.” They are not the same judgment.
An Oscar-winning, festival-recognized, or GAC-recognized film may be superb, and I may still have no desire to watch it tonight. Perhaps I just finished a long day of work and want something light. Perhaps I am researching a specific historical question and would rather watch a modestly produced but factually rigorous documentary. Perhaps I have only fifteen minutes instead of two hours. The same person also changes from moment to moment. A weekend evening is different from a flight, and a lunch break is different from a quiet Sunday afternoon.
A single universal ranking cannot fully capture that. Curation in an era of near-infinite media supply will therefore need two complementary systems. The first is public evaluation: what is excellent, meaningful, original, and culturally important? The second is personal matching: what is appropriate for this person, in this context, now?
This is not an argument against shared standards. Quite the opposite. Trusted public evaluation can become one of the most valuable inputs into personal matching. But it cannot replace the final contextual decision. “Is it good?” describes the relationship between a work and an evaluative standard. “Is it right for me?” describes the relationship between a work and a particular person in a particular situation. The more abundant content becomes, the more important both relationships become.
This is where another actor enters the picture: the Personal Agent, or PA.
5. A Personal Agent Is Not Just Another Recommendation Algorithm
Platforms already personalize. Netflix matches content using viewing behavior, feedback, and other signals. YouTube does not rely only on clicks and watch time; it also uses satisfaction-related signals. So it would be wrong to say that current recommendation systems simply do not care about users.
The deeper distinction between a PA and a platform recommendation system is whose agent it is and who gets to define success. A platform recommendation system belongs to the platform. The platform must balance user satisfaction, business revenue, content supply, advertiser interests, and its own rules. User interests may align with the platform’s interests, but they do not always align perfectly.
A Personal Agent, by contrast, should be owned and controlled by the individual. It represents the person in the media world. It may use recommendation algorithms, call external models, and consume public evaluation signals from critics, festivals, institutions, or organizations such as GAC. But its objectives, permissions, and evaluation criteria should ultimately come from its owner rather than from the content supplier.
I might tell my PA that tonight I want only one film recommendation because I do not want to spend half an hour choosing. I might tell it that this week I want to enter one unfamiliar field instead of seeing more of what I already like. I might also tell it that today I do not want to watch anything at all and should not be interrupted. For such an agent, helping me consume less media may count as success.
That is important because “belonging to me” cannot merely be a marketing claim. It should mean that I can understand why the agent made a decision, restrict what information it may use, revoke its permissions, and replace its provider when necessary. If a so-called Personal Agent primarily accepts hidden incentives from suppliers, then no matter how well it understands me, it is not truly my agent.
The most important change brought by PA is therefore not prediction but agency and delegation. A platform uses algorithms to compete for my attention; a PA should help me manage my relationship with the outside world. Both may be personalized, but the direction of power is different, and that difference may ultimately produce a very different media order.
6. AI Movie and Agentic Movie Are Not the Same Thing
Once we have a Personal Agent, the supply side also needs an entity that can interact with it. Let us call this a Media Agent. A Media Agent represents a work, a creator, or a content organization. It should be able to describe what a work is about, who it may suit, which versions exist, what uses are permitted, what the price is, and which conditions are negotiable.
The PA approaches from the other side, based on the person’s interests, available time, budget, and permissions. Suppose my PA finds a documentary highly relevant to me but also knows that I have only fifteen minutes. It might ask whether there is an authorized short version, whether there is a five-minute version that gives me a meaningful introduction, or whether the full version can wait until the weekend.
A five-minute version does not have to be a conventional trailer. It might be a genuinely coherent alternative presentation of the same underlying work. It is not equivalent to the long version, but it can be a different entrance into the same subject. Price may change accordingly. In a hypothetical transaction, I might pay twenty-five cents to watch the short version and then decide whether to purchase the fuller experience.
The important change is not simply that someone has cut a video shorter. The work begins to carry machine-readable descriptions of content, authorization, version relationships, and transaction conditions. It becomes something more than a fixed file waiting for a human click.
I use the term Agentic Media for a media system in which agents can participate in discovery, matching, negotiation, transaction, and presentation. When movies enter such a system, we can call them Agentic Movies.
The distinction is important. AI Movie primarily asks how a movie is produced, while Agentic Movie asks how a movie enters into a relationship with its audience. A film does not need to be AI-generated in order to become Agentic Media. Conversely, a film may be entirely AI-generated and still fail to change the basic media structure if all we do is upload it to an existing platform and wait for clicks.
7. You Cannot Invent Radio by Inventing Only the Radio Station
I collect two old vacuum-tube radios from roughly a century ago. Looking at those four-tube and six-tube machines, one is reminded of something easy to overlook today: broadcasting did not become a mass medium simply because someone built a broadcast station.
One side required studios, transmitters, modulation, power, and antennas. The middle required spectrum and transmission infrastructure. The other side required millions of receivers inside people’s homes. No matter how powerful the transmitter, without the receiver, broadcasting could not become a household medium.
Television required the same basic completeness: cameras and studios on one side, distribution networks in the middle, television sets on the other. The Web followed a related logic. Tim Berners-Lee did not merely create electronic documents; he built early servers and browsers and helped establish the protocols and representation standards connecting both ends.
A new medium does not emerge simply because a new kind of content can be generated. We must also ask how that content is transmitted, how it enters everyday life, and what capabilities exist at the receiving end.
From this perspective, AI video generation solves a very important problem on the production side, but it does not by itself complete the invention of a new medium. We are rapidly building more powerful “broadcast stations,” while the other end is only beginning to emerge.
The Personal Agent may become the “radio receiver” of the next media era. It does not necessarily need to be a new piece of hardware. It might live on a phone, a computer, or another device. What matters is not the shell but whether an individual finally has a media endpoint capable of acting on that individual’s behalf.
8. This New “Radio” Does More Than Receive
Traditional radios can tune stations. Televisions can switch channels. Internet users have always been able to search, comment, and create. So the distinction is not that people in the past were passive. The difference is that more and more active behavior can now be continuously delegated to an authorized agent.
A PA can first help me understand a work and then decide whether it is worth bringing to my attention. It does not need to “watch” all 10,000 movies in full. It can begin with metadata, trusted evaluations, machine-readable descriptions, and existing analysis, narrowing the field step by step and spending more computation only where necessary.
This is also where public evaluation can become especially useful. A trusted assessment from a film institution or evaluation framework can become one signal among many that a PA uses when deciding whether something deserves more of its owner’s attention.
More importantly, the PA can also express demand in the opposite direction. It might tell a Media Agent that its owner is interested in this topic but does not want excessive violence, is willing to pay but does not want to disclose personal identity, or has only twenty minutes and wants a version that preserves the central argument.
So what moves between the two ends of the media system is no longer just content. It also includes intent, constraints, authorization, and conditions. Traditionally, people leave feedback after they have watched something, through clicks, watch time, ratings, and comments. A PA may bring the person’s requirements into the matching process before viewing. It does not wait for the platform to infer me; under appropriate authorization, it can speak for me.
That makes Agentic Media a two-way, negotiated media system. But one limit remains essential: machines can process information on my behalf, but they cannot replace my lived experience. My PA can help choose a film, but it cannot be moved by the film for me. Its role is to reduce the burden before experience, not to remove experience itself from human life.
9. A New Medium Requires a New Technical Architecture
If these relationships are to work, we cannot simply add a chat box to today’s video player. At the architectural level, Agentic Media requires at least four interacting layers.
The first is the Generation / Production Layer. This layer creates stories, shots, sound, characters, scenes, and different versions. It can include traditional filming and human artistic production as well as generative AI. Its purpose is to answer how content can become possible, not to imply that machines must independently create everything.
The second is the A2A Transport & Transaction Layer. This layer supports discovery, identity, communication, authorization, task coordination, payment, rights management, fulfillment, and settlement between agents. Here, A2A refers broadly to Agent-to-Agent interaction. Emerging A2A protocols are already beginning to address communication and collaboration across agent systems, but a communication protocol is not the same thing as a complete media market. Copyright, commercial terms, legal responsibility, payments, and settlement will require additional mechanisms.
The third is the Presentation Layer, which decides how a work should appear under the constraints of the work itself and the user’s circumstances. It may be the original film, an authorized short version, another language, or an interactive explanation. Creator control must be protected, because some works support multiple presentations while others depend on rhythm, sequence, and structure that should not be broken apart. Agentic Media should enable negotiation, not force every work into infinite customization.
The fourth is the Personal Agent Layer, which sits closest to the human. It manages the preferences, context, privacy boundaries, time budget, money budget, and action authority granted by the owner. Which actions may the agent take autonomously, which require confirmation, and which types of content should never enter regardless of relevance are all questions that belong here.
These four layers are logical functions. They do not imply that four separate companies must provide them, nor do they form a one-way pipeline. Content flows toward the person, while human demand flows back through the PA into production and transaction. Trusted evaluation can enter the system as another important source of public knowledge. It need not become a fifth mandatory technical layer; it may function as an institutional signal or specialized service that Media Agents and Personal Agents can consult.
Identity, privacy, authorization, and auditability must run across the entire system. Otherwise, adding agents will not automatically increase personal sovereignty. It may simply create a less visible form of control. The technical details go much deeper, but the architectural point is enough for this essay: once production, connection, presentation, and the personal endpoint all change at the same time, the business model no longer needs to imitate the traditional film industry.
10. Why Produce All 10,000 Movies First?
Now we can return to the original thought experiment. The most straightforward approach would be to produce all 10,000 movies, release them all, and then use advertising, recommendation, evaluation, and social distribution to find audiences. But if both production and matching are becoming intelligent, why must we preserve that order?
This is where Kickstarter offers a useful analogy. A project can present a concept, prototype, and production plan before full-scale manufacturing. It can collect support and test whether demand exists. Film and other cultural projects can also seek support before completion. Pre-sales, crowdfunding, patronage, and commissioning all existed before AI, so the innovation is not that we can finally find an audience before making a movie.
The real change lies in who performs the search, how judgment is made, and how low the cost of participation can become. Traditionally, creators must continually promote themselves, while audiences must manually discover a project, read descriptions, assess trustworthiness, and decide whether to pay. Every step consumes human attention.
If PAs and Media Agents can absorb part of that work, demand matching before production may move from an occasional activity into a continuously operating mechanism. Perhaps the 10,000 things entering the market first do not need to be 10,000 finished movies. They could be 10,000 concepts worth exploring.
11. My PA Could Discover a Movie That Does Not Yet Exist
Take my own interests. I spend a great deal of time thinking about technology anthropology, and I am also interested in sociology, AI, and art. In the future, I could authorize my PA to participate in relevant thematic directories so that suitable creative projects can find it.
What becomes discoverable does not need to be my private life or my entire personal dataset. It could simply be an agent endpoint authorized to receive certain kinds of proposals. A producer might know that such a PA exists without having any right to know everything about the human behind it.
Suppose someone wants to make a documentary about human agency in the age of AI. The project is not finished; it has a creative plan, some sample material, and a budget. Its Media Agent discovers my PA, which evaluates whether the theme is relevant, whether the production team is credible, whether the promises are reasonable, and then decides what to do based on the authority I have given it.
I might pre-authorize a rule under which the PA can support projects up to one dollar autonomously, no more than five times per month. Five-dollar projects might require stronger evidence, while at twenty dollars it would have to ask me. The exact thresholds would be personal.
The point is that I no longer need to spend half an hour browsing a website, comparing projects, and filling out payment information for every one-dollar decision. My interests can participate in an authorized matching process even when I have not opened an app.
Of course, a PA saying that its owner may be interested is not yet real demand. Demand becomes economically more meaningful when that interest is expressed through a limited budget, a pre-order, a pledge, or another concrete commitment. Feedback that traditionally appears only after viewing can begin to become action before production.
12. I Could Even Give My PA a $100 Micro-Fund
As a venture investor, I naturally take the idea one step further. Why not let the PA manage a small cultural-project budget?
For example, I give it $100 and allow it to spend no more than five or ten dollars per month supporting documentaries, short films, art projects, or knowledge projects. I do not need every project to generate financial returns. I can tell the PA that one portion of the budget is for supporting work I want to exist, another portion is for purchasing future viewing rights, and only a smaller portion may be used for opportunities that include some economic participation.
When the works are completed, I, the physical human, still consume and evaluate them. I can tell my PA that a project chose the right subject but treated it too superficially, that another fell outside my normal interests but gave me an unexpected insight, or that a team overpromised and underdelivered and should face a higher diligence threshold next time.
Discovery, commitment, actual consumption, personal evaluation, and future selection can then form a continuous learning loop. The PA no longer learns only from what I clicked. With my permission, it can learn which experiences I later considered worthwhile, which were merely momentary temptations, and which unfamiliar experiences deserve more exploration.
It should not freeze me into a permanent user profile. I might deliberately set aside a small amount of money for serendipity and experiments I am not sure I will like. A good Personal Agent should not merely reinforce my existing preferences. It should also help me pursue forms of growth that I myself value.
This echoes another important point made by Herbert Simon: a system that truly saves human attention should absorb and process more information than it pushes outward to the user. A micro-fund managed by a PA is therefore not merely an automated payment tool; it could become a new way for individuals to participate in cultural production.
13. Why a One-Dollar Commitment Could Become Economically Meaningful
Traditional investing has transaction costs: sourcing, meetings, diligence, negotiation, legal documents, and ongoing management. Fund size, portfolio construction, risk distribution, and exit dynamics all influence investment strategy, and high processing costs make many small opportunities economically irrational for professional investors.
Small projects are not necessarily worthless. Their value may simply be too small to justify the cost of organizing the transaction. A2A may change that threshold. If portions of discovery, verification, matching, contracting, and settlement can be performed at very low cost, transactions that were previously too small to organize may become viable.
We do not need to assume that all costs go to zero. They only need to fall far enough to become compatible with micro-transactions. A highly specialized documentary might generate only $10,000 over its entire life. That may not fit the cost structure of a conventional film business, but if production, customer acquisition, and service costs are sufficiently low, such a project could still sustain itself.
One distinction is essential, however. One dollar can represent very different relationships. It may be a donation or pledge, a pre-order that creates delivery obligations, or an investment with a claim on future income. These are not legally or economically interchangeable. Kickstarter itself does not offer equity, and securities-like crowdfunding may fall under separate rules governing issuance, disclosure, and investor protection. Automating a transaction through agents does not make those obligations disappear.
Within those boundaries, however, an important possibility remains: AI may expand the supply of long-tail content, while A2A may reduce the cost of connecting long-tail capital to that content. Tiny expressions of support scattered across the world may become economically organizable in a new way.
14. My PA Does Not Need to Be an Expert in Everything
This raises an obvious problem: how does my PA know whether a production team is trustworthy? Liking an idea is not the same thing as believing a team can execute it. A beautiful sample reel is not the same thing as a credible production plan. If a PA simply treats promotional material as fact, it will merely automate human gullibility.
Agentic Media therefore needs specialization. A PA should not be expected to know everything. It may consult a third-party DD Agent, or diligence agent, to examine the team’s delivery history. It may consult a copyright service to examine chains of rights and source material, use reputation services to identify repeated delays or broken promises, and rely on payment and settlement agents to handle fulfillment conditions and revenue distribution after a transaction.
Trusted evaluation organizations can play another role here. They are not substitutes for diligence, but credible records of artistic and technical evaluation can become part of the broader information environment on which agents rely.
These services should rely, as much as possible, on verifiable records rather than on several models making the same guess and then calling consensus “cross-validation.” Conflicts of interest must also be visible. Otherwise, “independent diligence” may become merely another layer of the producer’s marketing.
There is also an important economic design principle here. We should not run five full diligence processes every time someone spends one dollar. Basic identity verification, delivery history, and rights information about a team can, under appropriate authorization and update mechanisms, be reused by many PAs. The expensive foundational verification can be shared, while personalized judgment is performed at the margin.
This is what may make professional services economically compatible with very small transactions. The significance of A2A therefore goes beyond AI systems talking to each other. It may allow professional work to be decomposed, reused, and recombined, shrinking the minimum efficient scale of organized economic activity.
15. From Producing First and Finding the Audience Later, to Letting Demand Participate in Production
Now we can connect the pieces. A creator proposes a concept, explains the goal of the work, provides necessary samples, evidence of capability, and a budget. The Media Agent finds potentially relevant PAs. The PAs perform interest matching and, where needed, call third-party services. Once sufficiently clear support, pre-orders, or commitments form, the project advances to the next stage.
Creative exploration does not disappear, nor does development work. The change is that we no longer need to complete the entire work before seriously testing its relationship with an audience.
Out of 10,000 concepts, perhaps only several hundred receive sufficient support. The others may reduce scope, revise their approach, seek a different audience, or pause. Production capacity remains powerful, but it no longer has to turn immediately into a mountain of finished content and then push the burden of finding value onto the audience.
I call this Demand-before-Production. More precisely, it means allowing demand to participate in resource allocation before full production, rather than waiting until after completion and using clicks as the final verdict.
This differs from ordinary market research. Answering “that sounds interesting” is not the same as making a trackable commitment within a limited budget. But financial commitments are not perfect forecasts either. People can make mistakes, projects can fail, and groups can be manipulated by the same promotional narrative. Commitment, delivery, and post-consumption evaluation must therefore remain connected.
At the same time, Demand-before-Production must not become a rigid rule forcing all creative work to obey immediate demand. Some of the most valuable work exists precisely because an artist or thinker saw something before an audience knew to ask for it. Exploratory work, personal expression, and public funding still matter. The purpose of bringing demand forward is not to let the market preemptively kill uncertain ideas, but to give support that already exists, yet is currently too fragmented to coordinate, a better chance of helping a work come into existence.
16. From Long-Tail Content to a Long-Tail Economy
We can go one step further. Demand does not always have to wait for a creator to propose a project first.
Suppose many people’s PAs independently discover that their owners are searching for a kind of content that does not yet exist: perhaps a documentary that discusses ordinary life in the age of AI without treating AI either as salvation or as apocalypse. Within their respective authorization boundaries, those needs could be aggregated and creators willing to respond could then be discovered.
People who do not know each other would not need to form a community, schedule meetings, and organize manually before learning that they share a similar desire. The Media Agent would no longer only market existing works; it could also discover an unmet subject worth creating.
Creators would not have to obey demand literally. They could offer their own interpretation, and the two sides could gradually develop a viable project. Personalization also does not mean that every person needs a completely separate movie. A single work can preserve a common narrative core while offering different languages, lengths, or entry points to different people. That preserves shared production economics and also preserves something culturally important: a common object that people can still watch and discuss together.
A long-tail economy therefore means more than storing lots of niche content on a platform. It means that niche demand can find creators, obtain support, complete production, reach an audience, and feed its evaluation into the next round of activity.
A cultural market with only a few thousand viewers may not support a large institution, but it may still support a small team. The key question is not whether the audience is “mass” enough; it is whether the project can find the people who truly need it at an economically appropriate cost. The niche stops being merely the leftovers of a mass market and can become a relatively complete small market of its own.
17. A New Agentic Media Economy
At this point, Agentic Movie is clearly more than using AI to make filmmaking cheaper. Several changes begin to reinforce one another: production technology makes smaller-scale creation viable, trusted evaluation helps society distinguish excellence and meaning from pure volume, PA makes it easier for individuals to express and execute preferences, A2A helps fragmented supply and demand discover each other, professional services provide trust support for micro-transactions, and the Presentation Layer allows the same work to enter different people’s lives in appropriate ways.
I call the economic system formed around these relationships the Agentic Media Economy.
In this system, the audience no longer appears only at the end of the value chain. A person may be a pre-order customer, an early supporter, or, under appropriate legal arrangements, a micro-investor. The creator may gain not only post-release revenue but also pre-production evidence of demand and commitments of resources.
Business opportunities therefore extend far beyond the generative model itself. Someone must help creators describe projects accurately, establish trusted evaluation and recognition, maintain reliable information about works, rights, and provenance, provide low-cost fulfillment, payment, and settlement, and help PAs truly understand and serve their owners.
This does not mean platforms disappear. Platforms can still provide aggregation, hosting, transaction, discovery, and reputation services. Independent institutions can provide public evaluation and cultural recognition. Personal Agents can represent individual demand. Different actors can perform different functions without one platform having to own the entire relationship.
As an investor, one possibility particularly interests me: an individual cultural project may be small, while the market for infrastructure serving millions of such small projects may be very large. We do not require every documentary to become a blockbuster, just as we do not require every small business on the internet to become a giant technology company.
Historically, media business has largely been framed as a competition for more attention. Another kind of business may emerge: helping limited attention and limited budgets meet the right creative work, and earning value by reducing the cost of that match.
18. We Are Inventing a Medium That Has Never Existed Before
Let us return to the opening question. If 10,000 movies are born every day, what do we actually need?
More generation may give us richer supply, but it can also push an even larger burden of choice onto people. Trusted evaluation becomes indispensable because abundance without discernment easily becomes noise. But evaluation alone cannot determine what every individual should watch, support, or help bring into existence. Better recommendation may improve matching, but it does not necessarily change who defines the objective, who owns the data, or who sets the transaction rules.
The deeper opportunity is to redesign content production, public evaluation, agent collaboration, personal agency, and resource allocation as parts of a larger media ecosystem.
Every component has historical precedents. Humans have sponsored creative work for centuries. Works have long been pre-sold. Film institutions have long evaluated excellence. Media has long collected audience feedback. Consumers have long used intermediaries and representatives. What may be genuinely new is the possibility of making all these relationships operate repeatedly, at low cost, across vast numbers of ordinary people, tiny budgets, and continuously changing content.
A century ago, the broadcast station and the radio receiver together created a new medium. Today, generative AI is transforming the production side. Institutions such as Global AI Cinema are beginning to address the public problem of trusted evaluation in an age of abundance, while Personal Agents may transform the human side.
This new “radio,” however, is different. It does not merely receive a finished program. It can represent its owner in searching for works that do not yet exist, state requirements, commit support, and bring the owner’s real experience back into the next round of creation.
The movie therefore no longer has to remain a fixed product that is completed first, distributed second, and then waits for an audience. It can begin as a concept, establish relationships with potential viewers, and gradually acquire resources and form under the leadership of creators. Human demand no longer needs to appear only as statistics after release; it can enter the creative process earlier as authorized action.
Seen this way, the AI cinema era is producing three increasingly important questions. AI generation asks what can be made. Trusted evaluation asks what is excellent, meaningful, and worthy of recognition. Agentic Media asks what can find the right people, and what those people might in turn help bring into existence. All three matter. The first expands production, the second protects discernment and cultural value, and the third may create a new relationship between creators, audiences, capital, and media itself.
That, to me, is the most exciting implication of the “10,000 movies a day” thought experiment. The future worth anticipating is not simply a world with vastly more films. It is a world in which excellent work can be recognized, individuals can retain agency over what deserves their attention, and creative projects that were previously impossible because audiences were too fragmented, budgets too small, or coordination costs too high can finally find the people who want them to exist.
When that happens, we will not merely have invented a more efficient way to make movies. We will have invented a new kind of media—one in which human judgment, individual agency, creative imagination, and social resources can continuously find one another.
About the Author
Chun Xia, Ph.D., is a co-founder of Silicon Valley deep-tech venture capital firm TSVC, the first fund invested in Zoom. He holds a bachelor’s degree in electronics and a master’s degree in computer science from Tsinghua University, and a Ph.D. in computer science from the University of Illinois. A serial entrepreneur who founded three technology companies, he previously served as a chief architect at Sun Microsystems. He is also a researcher in techno-anthropology and humanity AI.