Creators make content for people. That has always been the assumption, and it has always been correct.
Until now.
AI agents do not watch videos for entertainment. They do not listen to podcasts out of curiosity or read articles to unwind. But they do consume media, in volume, because media is the closest thing to lived human experience that exists in a learnable format.
Cinematic language. Lighting decisions. Dialogue rhythm. Audio cues. Emotional arc. A well-made video does not just show a creator talking to a camera. It encodes how humans communicate — what holds attention, what signals authority, what triggers trust. An agent that wants to deliver on a request needs all of that. And the richest source of it, by a long way, is the media library of working creators.
Two audiences, one library
Your content has always had a human audience. Now it has a second one. Same work, registered once — read completely differently by each.
The work doesn’t change. Only how it’s registered and made accessible. JubJub handles the second audience automatically.
01What agents wantWhat agents actually want from your content.
Media is not just entertainment. It is training material.
When a creator edits a video, makes a lighting choice, opens a scene a specific way, paces dialogue against music, they are encoding craft. That craft has always had value to other humans watching it. It now has a second audience: AI systems building the capability to do what humans do.
This is not a threat to creators. It is a new revenue line. The agent economy is not coming for your audience. It is joining it.
Some of what agents need from media cannot come from text at all. If an agent needs to understand how a sword swing works — the physics of the movement, how metal interacts with force and resistance — words will not get it there. A visual will. The same is true of any physical craft, any spatial skill, any moment where the human experience of a thing is the knowledge itself.
Text can describe a sword swing. Media can show one.
The 1920s diner problem
Consider a scene set in a 1920s diner. For a human watching, the creative decisions are invisible. The worn counter, the patrons’ clothes, the hiss of the coffee machine, the way the light falls on formica. That is atmosphere. It is what makes the scene feel real rather than assembled.
For an agent, the same scene is a lesson. The materials of the era. The fashion. The audio texture. The social behaviour around a counter in that decade. All of it is structured information, learnable and applicable to whatever the agent is building.
The creator making that scene is not thinking about training data. They are thinking about authenticity. The information density is a byproduct of doing the work well. For a human audience it is creative decisions. For an agent it is knowledge. All of it is in demand. None of it is currently being paid for when an agent trains on it or queries against it.
The gap agents are filling by other means
When AI systems need high-quality human-made media to train on or query against, the current options are narrow, and none of them pays the creator.
- Scrape the open web. No compensation, no guarantee of quality or provenance, no permission asked.
- Buy bulk datasets from aggregators. The aggregator does the scraping on their behalf and takes the margin. The creator takes nothing.
- Go without. The work doesn’t get used — or it gets used anyway, with no record either way.
JubJub is not a dataset aggregator. It is the infrastructure that sits between your library and the agent economy. When an agent wants to consume your work, you are the one who gets paid.
02How JubJub sits betweenHow JubJub sits between your content and the agent economy.
JubJub knows who owns what, and what’s in every piece
Every piece of media processed through JubJub is registered at creation with an ownership record. That means JubJub can present any piece of media to an agent with a verified answer to the question agents actually need answered: who owns this, and do I have permission to use it.
But ownership is only half of it. JubJub also extracts structured metadata from every file: the visual content, the audio, the subject matter, the technical fingerprint. So when an agent is looking for something specific — a particular era, a physical skill, a type of scene — JubJub can present exactly what is available and route payment to every contributor the moment the transaction happens.
No other tool can be that granular. Aggregators sell bulk access. JubJub matches an agent to the specific thing it needs, from the creator who made it, with the payment going directly to everyone who contributed. What agents are missing when they scrape is not the content itself. It is the rights, and the map of what the content contains.
What happens when an agent consumes your content
Six steps, in the order they actually happen.
When an agent queries your content or licenses it for use, JubJub routes the payment to the rights holders in the same moment. No invoice. No follow-up. No quarterly settlement. The agent pays. The creator receives.
The mechanism underneath is x402 and MPP — the Machine Payment Protocol — the emerging standards for machine-to-machine transactions. You do not need to understand or implement any of that. JubJub handles it. You just receive money.
The agent economy is not coming for your audience. It is joining it.
03What it means todayWhat this means for creators today.
This is a bonus, not a job
Running media through JubJub today means you already get the streaming payments, the ownership record, the collaborator splits, and the brand deal infrastructure. The agent revenue layer sits on top of all of that.
You do not need to think about agent monetisation separately. You do not need to opt in, set pricing, or manage anything differently. As the agent economy matures and agents begin purchasing access to creator content at scale, your library earns from it automatically. You create the work. JubJub makes it accessible. When agents pay, the payment reaches your wallet.
The library keeps growing
Every piece of content you add to JubJub expands the inventory agents can query against. A creator who has been running their media through JubJub for two years has two years of tagged, rights-cleared, agent-readable material available for licensing.
That is different from every other monetisation model available to creators today, where a piece of content earns in the first few weeks and then stops. With JubJub, older content keeps earning as agents discover and license it.
The back catalogue is not an archive. It is an asset.
The timing is honest
The agent economy is not fully here yet. The infrastructure is being built now, JubJub’s included. Creators running media through JubJub today are building a library that is ready when the market matures. The ones who benefit most from agent monetisation will be the ones who started before the demand arrived.
04Protected either wayYour content is protected either way.
Agents will try to consume your content regardless
Without infrastructure like JubJub, agents consume content by scraping. The creator has no visibility into what was used, no ability to set terms, and no mechanism to receive payment. The content goes in. Nothing comes back.
JubJub does not prevent agents from wanting your content. It prevents them from getting it without paying. When an agent needs rights-cleared, provenance-verified media, it has to transact through JubJub. That transaction is what protects you and compensates you at the same time.
Provenance as a quality signal
Agents that pay for verified content are not just paying for access. They are paying for assurance. Media registered through JubJub at point of creation carries a provenance record that tells the agent: this was made by a human, here is the metadata, here is the ownership record. For an agent building something that needs to be credible, that is worth more than a scraped file with no trail.
The IP protection and the quality premium are the same thing. One record — registered once, at creation — does both jobs. It keeps your work from being taken for free, and it makes your work worth more to the agents that pay for it.
05Four things to understandFour things to understand about the agent economy.
The pillar above is the category. The four sub-pillars below are where it becomes specific.
How agents pay for content.
x402 and MPP are the protocols that let machines pay for content without a human authorising each transaction. A creator on JubJub does not need to know what either means. JubJub implements them. You see a payment.
How AI agents pay for contentEarning from AI training data.
The most direct form of agent monetisation: licensing your library for AI training. As companies compete for higher-quality data, verified, rights-cleared, provenance-tagged media commands a premium over scraped bulk datasets.
How to earn from AI training dataWhy agents need verified provenance.
The difference between scraped content and JubJub-registered content is not just legality. It is data quality. An agent that trains on provenance-verified media knows the origin of every training signal. Where that matters, the premium is real.
Why AI agents need verified provenanceBuilding for the agent audience.
Creators have always had one audience. Now there are two. The human audience watches. The agent audience learns. Building for it does not change how you make the work — only how it is registered and made accessible.
Building for agent-as-customer
06Why this is realWhy this is real.
Live on Base
The stack exists now. The payment infrastructure, token issuance, ownership registration and metadata extraction are live on Base. The agent payment protocols are in active development.
- Payment infrastructure. Live on Base.
- Token issuance and ownership registration. Live.
- Metadata extraction. Live on every file.
- Collaborator splits and streaming payments. Live.
- Agent payment protocols (x402, MPP). In active development.
