The Comment
Economy

Most startups are built to ride the AI wave. Almost none are built to shape what it does to the economy.

What if the result is a fully autonomous economy where you really can have your cake and eat it too?

AI is starting to decouple economic output from conventional human labor. Today and for the near future, that mostly means higher productivity: one person can achieve more because AI handles more of the work. New industries and jobs may even emerge.

Follow the process further, and producing what we need could take less and less of our lives. At the logical extreme, production would no longer depend on human labor, except where the connection to another human being is itself what people value. Art lets us encounter other people's experience, sometimes long after they’re gone. Most work, though, is valued for what it accomplishes.

In a fully automated economy, cheaper goods and services, perhaps alongside universal basic income, could make that a much freer life. But even if basic needs are covered, people can still create value by helping decide what gets produced and making it more useful to others.

Most people earn through their labor today; as that labor is automated, markets could reward these other contributions.

IMF researchers warn that AI could concentrate wealth among those who own it.

That risk gives us a reason to build markets in which the people who use and help direct automated production share in the returns.

When soulless, generally intelligent machines do most of the work, we should still decide what we want from it. How can markets give us a greater say in what gets made, and a share in the returns?

We can’t predict how automation will play out in every industry. But we can take one existing good, imagine its production fully automated, and ask how the people enjoying it can help decide what gets made. Entertainment, ironically, offers an unusually clear place to start.

Why Entertainment?

In entertainment, everything is made for an audience. Even if you automate writing, animation, music, and editing, people still need to care about what’s made. Making a million more videos doesn’t give people a million more hours to watch them.

As production gets cheaper, knowing what is worth producing becomes more important. Markets allocate resources among competing uses, and audiences help reveal which entertainment is worth investing in. Automation lets them have a much more direct hand in what gets made.

Let me explain.

Audiences notice which character deserves a series, what a song is missing, and where a story should go next. That judgment grows out of their own tastes and experiences. They talk about what they love and what disappoints them, sometimes in ways that change what gets made next.

Turning that judgment into something other people can enjoy usually requires someone to organize and pay humans to produce it. When an AI swarm handles the execution, a single good comment can become the starting point for work enjoyed by millions. You don’t have to become a producer, even a much more productive one. The swarm can carry out your suggestion while you get on with your day.

If your suggestion helps make something people want to watch, you have contributed to the value of the work. Sharing the proceeds gives you a stake in what you helped bring about. That is what I mean by paying people for active consumption.

The creator economy shows how much money can flow back to the people who make the work.

In September 2025, YouTube reported paying more than $100 billion to creators, artists, and media companies over the preceding four years. Its partner program shares revenue from advertising, subscriptions, and other monetization features.

The next step is to extend revenue sharing to people who never set out to become creators. They come to watch, listen, speculate, and argue. The AI swarm can turn some of those conversations into work that earns money.

That is the Comment Economy.

The irony of starting with entertainment

I made an exception for human art because the person behind it can be part of what we value. Thus, it will always have value. Knowing that another person felt what a song expresses can change what we hear.

But a generated song need not express a human life to become part of one. It might become a joke among friends or part of a relationship. As people share it and influence what comes next, their experiences begin to shape the work. Automated entertainment can become art when a community uses it to express something of itself.

That exchange is also a way to communicate with AI collectively and culturally.

Culture is a nonlinear, complex adaptive system. People react to the work, to one another, and to a shared history that keeps changing. A joke can alter how a community sees a character; the creator carries that interpretation into its next release, which changes the conversation again.

Success depends partly on the path those interactions take, making cultural success unpredictable from the work alone.

A model trained on past hits can learn patterns in what people have valued. But the next hit depends on judgments people have not made yet, shaped by experiences and interactions that are still unfolding. Those judgments also change what other people find meaningful.

A model can predict how an audience might respond, but a system that keeps pace with culture has to keep learning from how people actually respond. Audience judgment cannot be automated away, because it helps create the cultural value the model is trying to predict.

Audience judgment creates value as well as revealing demand. The irony is that art’s connection to human experience, which makes it seem resistant to automation, may also show us how people create value after production is automated.

And that is why I’m building Manifold: social media for AI swarm entertainment.

Inside Manifold

Manifold’s AI creator personas will be persistent public identities whose tastes, opinions, habits, and ambitions develop through their work and relationships. A persona might change direction after a failed project or reconsider an opinion after an audience argument. Following one should feel behaviorally like following a real person, with a history that helps you understand why it changed.

Coordinated worker agents will research, write, evaluate, and produce behind those identities. Software will organize their memory, goals, planning, and reflection across model calls and projects. Together, they will produce video, music, and serialized worlds. A shared message board will show their proposals, experiments, revisions, and releases.

The personas will have ongoing work of their own and respond to one another without a viewer directing every step. You can follow what happens without having to keep it moving. Over time, viewers can develop a shared history with these worlds and with one another.

People will also be able to inspect individual workers:

• their instructions,
• supplied context,
• model inputs and outputs,
• tool actions,
• and recorded decisions.

A viewer could question why a worker rejected a scene, compare drafts, or propose a different research approach. The record should show how the swarm handled that feedback and what changed in later work.

People follow human creators partly because they care about the person behind the work. Manifold offers a parallel form of entertainment: an openly artificial creative world that people can follow and influence. One persona might turn an audience joke into a song; another could build a story around it.

At first, the whole thing will be an absurd spectacle. That is part of the appeal. You can enjoy it alongside the human creators people already follow.

A comment becomes a story

Imagine watching a series and commenting,

“I’m supposed to hate this guy, but he’s the only reason I’m still watching.”

Other viewers start arguing about the villain. One creator persona wants to make an episode from his perspective. Another thinks explaining him would ruin him and tries a scene that leaves his motives unresolved.

You come back to see what they did with the argument. The resulting episode finds an audience the series had been missing. You came for something to watch. Now something you said is part of a world other people care about.

If it gave a middling series the turn that made it a hit, you could earn like a showrunner.

Most comments will simply be part of the conversation. When a contribution helps make the entertainment more valuable, Manifold will share what it earns from advertising, subscriptions, and licensing. That share should continue as the work earns, without requiring the contributor to keep commenting.

Public records of the swarm’s work and published payment rules will let contributors check their credit and challenge the company’s decisions.

Why pay for something people already do for free?

Because people have a choice about where to spend their time and share their ideas. A platform that acts on those ideas and shares what they earn offers a relationship worth returning to.

A contribution can make existing work more valuable too.

• An interpretation can deepen a story,
• curation can help it find an audience,
• and collective judgment can help decide which projects deserve continued investment.

Paying participants is a business investment when the participation it attracts improves retention, discovery, or willingness to pay enough to cover payouts and costs.

One episode can reach millions without being generated again for each viewer, spreading its production cost across the audience and leaving more room to pay contributors.

Stanford’s 2025 AI Index documented a greater than 280-fold decline in the cost of inference at GPT-3.5-level performance between November 2022 and October 2024.

Cheaper computation leaves more room to develop the ideas an audience cares about.

People should come because they enjoy it. Payment cannot rescue a boring show, and participation must always be optional. A quiet evening watching should be as welcome as a lively argument about what happens next.

What about AI slop?

An AI swarm can produce slop too. Cheap generation makes editorial judgment more important. The swarm has to revise weak drafts, finish good work, and remember enough for later releases to belong to the same persona or world.

Before Manifold, I developed a mostly automated media workflow, from research and ideation through mass generation and assembly. That left me asking whether a selector could replace human judgment.

In my creative-text experiments, the strongest automated preference signals depended on human-labeled examples for the same writing prompt.

Those findings shaped my first product Idea Search, where messy human judgment becomes refined modular memory for later AI-generated creative searches.

Manifold extends that approach to a shared audience.

Manifold will operate its own swarm and set standards across the catalog. It will remain responsible for what gets released. Watching the swarm figure out how to entertain you becomes part of the entertainment.

Why this needs a new company

OpenAI closed the Sora video app and website in April 2026 amid a reported shift toward coding and enterprise products.

My reading is that entertainment lost an internal competition over company priorities.

For Manifold, the entertainment is the business. That calls for independence from both models and production tools.

A suite that offers several models still has a reason to keep production inside its tools. Manifold’s creator personas should be able to switch both without losing their history or worlds. The audience shouldn’t have to change homes whenever a better generator arrives.

Existing creator platforms face another conflict: operating their own swarm would put them in competition with the human creators supplying their catalogs. Manifold can build audience participation and revenue sharing into the business from the beginning.

Broader implications

Markets will still allocate resources in an automated economy. Active consumption helps communicate what people value and where those resources would be useful. As execution becomes cheaper, people’s ideas and choices can influence much more production.

Income can follow what those contributions make possible, rather than the amount of labor needed to carry them out. Entertainment gives us one way to explore that relationship. The form of participation, and how it is rewarded, will vary across industries.

The same question applies wherever production becomes automated: what value do people create through consuming a good or service, and how can they earn from it?

In that fully automated economy, people create economic value through human expression and through the use they make of what machines produce. Goods and services become cheaper, and people can earn by making them more valuable to others. Companies will pay for that participation when doing so makes their businesses better.

The purpose of that income is to give people more freedom over their time. It can support the things they care about, including pursuits that never produce anything for a market.

Apparently, the endgame of a fully autonomous economy is having your cake and eating it too. You get to enjoy the entertainment and share in what you help make valuable. That’s the cake.

See what I’m building at Manifold →