When content can become an App, an agent, or an interactive world, the platform around it has to change too.
In 1977, Alan Kay and Adele Goldberg published a paper called Personal Dynamic Media. The personal computer was still an unfamiliar object, but they were already arguing that it should be understood as more than a faster typewriter or a smaller television. They called it a "metamedium": a medium capable of containing a wide range of existing forms of expression, as well as forms that had not yet been invented.
Nearly fifty years later, the computer has absorbed almost every medium we knew. We read on it, watch on it, draw on it, compose music on it, and build worlds inside it. Yet most software still arrives in familiar containers. A document resembles a page. A video sits inside a player. An App opens into a sequence of screens. AI, despite all its apparent intelligence, first met most people through one of the simplest containers of all: an empty chat box. Now those containers are beginning to loosen.
The great content platforms of the internet did more than host files. Blogs made it ordinary for a person to publish writing without owning a printing press. YouTube organized online video around channels, subscriptions, comments, and recommendations. Instagram made the mobile image a social object, while TikTok built a culture around short video, sounds, templates, and Duets. None of these platforms invented the medium at its center. Each made a form of creation easier to publish, easier to encounter, and easier for an audience to respond to.
The AI era presents a less settled object. Its native content might be an answer, an image, an App, a character with memory, or a world that responds while someone moves through it. If the content itself can act, adapt, and become the starting point for another creation, what kind of platform belongs around it?
On most earlier content platforms, interaction gathered around the work. A person could like it, comment on it, subscribe to its creator, or pass it along. On a platform for playable work, action can also happen inside the work. A person can use an App, affect an Agent, explore a World, and then Remix the experience into something new. The social layer does not disappear; the content itself becomes another surface of participation.
One Eazo demonstration begins with a toy sitting on a table. Someone takes a photograph of it, and the familiar object is carried into a digital world. There, it can have an identity, memory, and behavior. A wave can bring it closer; a raised hand can make it stop. What has been made: an image, an App, a character, an Agent, a game, or a small world?
Sometimes a thing is difficult to name because it has been poorly defined. At other times, the nouns begin to fail because the medium itself is changing. The question for the AI era is therefore not only what new forms software can take, but how those forms will be created, discovered, experienced, and continued between people.
The Computer Could Contain Any Medium, but Most People Could Not Shape It
Kay and Goldberg imagined the computer as a medium that could simulate other media and make room for new ones. In that sense, their prediction came true. A laptop can be a book, a cinema, a recording studio, a drafting table, a laboratory, and a place to play. Software can combine text, sound, image, animation, calculation, simulation, and communication in ways no physical medium could comfortably contain.
But this freedom has never been shared equally between the people using computers and the people deciding what computers can become. Most of us encounter software through forms someone else has already completed. We can fill in the fields, choose among the settings, and move through the paths provided to us. Inventing the paths themselves has required another kind of access: knowledge of programming languages, interface design, data systems, infrastructure, and the long process of turning all of them into a dependable product.
The computer became a metamedium for its users before it became one for most of its authors. The machine could theoretically take almost any form, while the practical ability to give it a new form remained concentrated among people with the time, training, and resources to build software. Many ideas therefore entered the computer only after they had been translated into a familiar product category. Other ideas never entered at all.
The rise of vibe coding is one visible sign that this arrangement is shifting. The phrase is imprecise, and its most careless version confuses a generated first draft with trustworthy software. But it has caught on because it describes a new expectation: people can begin with an intention expressed in ordinary language, see software start to take shape, and revise it through conversation. They no longer always have to begin at the point where the machine's vocabulary takes over from their own.
Natural language does more than shorten implementation. It allows software creation to begin while an idea is still a story, an image, a situation, or a desired behavior. "Make a tracker" is a specification of sorts. "I want my daughter to photograph an animal and hear a story in which it recognizes her" is something less tidy and more revealing. It contains a participant, a moment, a relationship, and a feeling the experience should preserve. A broader software medium begins by allowing more of that original imagination to survive the journey into a working form.
When AI Leaves the Chat Window
Conversation was a natural first home for generative AI. Language models produce language, so we placed them in a box where language goes in and language comes back. The chat window made an unfamiliar intelligence legible: it behaved enough like messaging that a person could understand what to do before they understood anything about the model underneath.
Chat remains powerful, but it is a narrow stage. An answer appears and waits to be read. Even when the system remembers the conversation, most of its apparent world lies behind the text. It has no necessary position in space, no objects to act upon, no surrounding characters to notice, and no unfolding situation beyond the one described in words.
In 2023, a group of researchers from Stanford and Google offered a glimpse of what changes when an AI system is placed inside an environment. In Generative Agents: Interactive Simulacra of Human Behavior, they populated a small virtual town with 25 software agents. The agents kept records of their experiences, reflected on those memories, formed plans, noticed one another, and changed their behavior as the day unfolded.
The most memorable moment began with one instruction: an agent wanted to hold a Valentine's Day party. Over the next two simulated days, news of the party spread through the town. Agents invited one another, made new acquaintances, asked each other on dates, and coordinated their plans so they could arrive together. The event was modest and the town was a research environment, not a functioning autonomous society. What mattered was the change in structure. A reply had become an event, and the event acquired consequences because the agents occupied the same place and time.
A chatbot waits at the edge of a situation. An Agent inside a world can become part of the situation. It may remember what has happened there, act among other things, and encounter consequences that were not contained in a single response. This does not make every Agent a person, nor does it make simulated memory equivalent to human memory. It gives creators a different expressive material: not only words an intelligence can produce, but behavior an experience can organize.
The toy in Eazo's demonstration becomes interesting at precisely this threshold. Turning it into a 3D model would change its representation. Giving it identity, memory, and actions changes the relationship someone can have with it. The object is no longer only shown. It can meet a person inside an experience.
A World Model Predicts What Happens After an Action
A world model is not simply another name for a video generator. In AI research, it generally means a learned internal representation of an environment: what state it is in, how that state may change, and what is likely to happen after an action. In their 2018 paper World Models, David Ha and Jürgen Schmidhuber trained a model to learn a compressed spatial and temporal representation of an environment, then let an agent practice inside what they called its own "hallucinated dream." A world model does not have to render a photorealistic scene. Its essential job is to make enough of the world's dynamics predictable for an agent to plan or act within them.
Video generation has made this older idea newly visible. OpenAI introduced Sora in a technical report titled Video Generation Models as World Simulators, pointing to the way a model can preserve aspects of space, objects, and motion as it extends a scene through time. The phrase also exposes the limit. Producing a convincing continuation is not the same as possessing a complete theory of physics, and the moments when an object disappears or an impossible motion looks briefly plausible reveal how partial that learned model still is.
Systems such as Google DeepMind's Genie move the idea closer to playable media by making a person's action part of what gets generated next. DeepMind describes Genie 3 as a model that can create dynamic environments from text and allow them to be navigated in real time at 24 frames per second, with consistency lasting a few minutes at 720p in its published demonstrations. In a generated video, the viewer cannot change the next event. In an interactive world, the participant enters the generation loop.
If someone moves left, the world must have a left side. If they turn around, something has to remain behind them. If they push an object, cross a stream, or return to a place they visited before, the environment has to preserve enough space, state, and causality for the action to make sense. A world is not simply an image with depth. It is an arrangement of possibilities that must continue answering the participant.
Today's generated worlds are still bounded experiments. They can lose consistency, forget objects, improvise unreliable physics, and offer far less precise control than a designed game or simulation. Calling them a signal does not require pretending those problems have been solved. The signal is that generative media are beginning to move from producing artifacts toward producing environments.
This distinction also protects the idea of a world from becoming a decorative fashion. A calculator does not become more meaningful because it sits in a mountain valley. A calendar does not need a character walking through a room. A conversation does not become more intelligent because the speaker has a 3D body. New forms matter when they fit the kind of participation an idea requires, not when every idea is forced into the newest form available.
The important shift is therefore not from Apps to Worlds, as though one were replacing the other. It is from software as a predetermined container to software as a widening range of expressive forms. An idea might need the clarity of a tool, the behavior of an Agent, the spatial relationships of a World, or a combination for which we do not yet have a settled name.
Interactive Work Needs Discovery, Not Just Creation
A medium needs more than a way to be made, and a platform does more than provide storage. It establishes habits around how people find a work, how they recognize its boundaries, and what they believe they can do when it arrives. The audience has already been learning some of the habits of playable work in places that are usually described as games.
Roblox is distributed as one platform, but what people encounter inside it is a changing field of experiences made by its community: obstacle courses, racing games, social spaces, shops, simulations, and forms that borrow from several categories at once. In the second quarter of 2025, Roblox reported 111.8 million average daily active users. The number does not tell us that all software will become a 3D game. It shows, at a scale difficult to dismiss, that people already understand how to enter a platform in order to discover many different interactive worlds.
That behavior differs from the traditional relationship between a person and an App Store. The App Store begins with a relatively stable object. A person identifies a need, searches a category, evaluates a product, downloads it, and returns when the need arises. A platform of experiences permits a more open kind of arrival. A person can enter without knowing exactly what they want, move from one experience to another, and understand each work by trying it rather than by reading a description from outside.
This is one reason Eazo places Apps, Worlds, Agents, and other interactive experiences in a Community rather than treating creation as the end of the product. A person might come to make something, but they might also come simply to see what other people have made. In the Feed, a working App can be encountered before someone has formed the search query that would have led to it. Software begins to participate in discovery, not only utility.
The distinction matters for creators as much as it does for audiences. An idea does not become culturally meaningful merely because it can be generated. It needs somewhere to travel, a way to reach people who were not present when it was made, and an invitation clear enough for them to enter. Creation tools can increase the supply of software. A medium also requires circulation.
Scratch Made Remix Part of Creation
Circulation changes again when the person who encounters a work can continue it. Long before generative AI, Scratch made this possibility part of its basic social grammar. A person could open an interactive story, animation, or game, look inside the project, change it, and share the result as a Remix. The platform preserved a visible relationship between the new version and the work it came from, and its Remix Trees showed how one project could branch through a community.
By June 2024, the Scratch Wiki estimated that roughly a quarter of recently shared projects were Remixes. The number is less important than the behavior it represents. For a generation of young creators, beginning with someone else's project was not treated as a failure to have an original idea. It was a normal way to learn, respond, and discover where one's own decisions began.
This is one of the clearest signs that software can behave like a cultural medium. A song can be sampled. A story can be retold. An image can be quoted, transformed, or placed in a new context. Software has always been copied and forked at the level of code, but that form of continuation has mostly belonged to people who can understand the underlying implementation. Remix brings the invitation closer to the experience itself.
On Eazo, when a creator enables Remix, a person who discovers an App or World does not have to return to a blank page. They can begin with the work that moved them, keep the structure that is still useful, and change what another context requires. A travel companion made for one city can be adapted for another. A character can enter a different story. A tool made around one household's routines can reveal a shape that another family recognizes, then become specific again in someone else's hands.
Remix is not the same as frictionless copying. A healthy practice depends on permission, attribution, and the creator's ability to decide whether and how a work can be continued. But when those relationships are built into the product, publication gains another possible future. A work can remain complete in itself while also becoming material for something it could not have anticipated.
Eazo Brings Creation, Discovery, Use, and Remix Together
None of these developments began in one place. Kay and Goldberg described the computer as a metamedium decades before natural-language creation was practical. Research agents showed how memory, planning, and environment could turn generated language into situated behavior. World models began turning prompts into places that answer movement. Roblox made worlds discoverable at enormous scale. Scratch made continuation visible through Remix.
What is new is not any one signal, but their convergence. The means of creation, the shapes software can take, the ways it can be discovered, and the relationship between an original work and its continuation are all changing at once. Together, they suggest something larger than faster development or a new category in an App Store. They suggest the need for a content platform built around software as an accessible medium for expression.
At Eazo, this is the kind of content platform we hope to build for the AI era. In Create, a person can begin with natural language or an image and work toward a functioning App, World, Agent, Skill, or interactive experience. The aim is not only to generate a plausible screen, but to carry an idea through the less visible parts of software creation until it can be opened and used. From there, the work can be published to the Community, discovered in the Feed, shared, and, when the creator allows it, Remixed into another version.
The path is simple enough to write in one line: Idea → Create → Publish → Discover → Use → Share → Remix. Its significance lies in bringing activities that software has usually separated into one creative loop. A generation tool can help make an object. A medium also needs ways for that object to travel, meet people, acquire meaning through use, and return as the beginning of another act of creation.
Our hope is not that every App should become a World, that every AI should be given a body, or that every person should call themselves a developer. It is that creators should have access to a wider expressive range, and that the platform should be able to carry those different forms from imagination into public life. A useful tool, a responsive character, an interactive story, and a place someone can enter do not have to be competing visions of software. They can be different answers to the same question: what form does this idea need in order for another person to take part in it?
This is also why the boundaries between Apps, Worlds, Agents, and Skills should not be made more rigid than the work itself. A World can contain Agents. An App can use a Skill. An Agent can become part of a tool, a story, or a place. The medium is not defined by the appearance of any single object. It is defined by what people can create around it, what others can do inside it, and whether the work can continue to generate new possibilities after it is published.
The Nouns Can Come Later
Kay and Goldberg were right that the computer could contain media that had not yet been invented. What remained scarce was not only computing power. It was authorship: the ability for more people to decide what form a programmable idea should take, and the ability for that form to travel between people.
The App is not disappearing. Neither are the page, the video, or the chat window. New media rarely erase the forms that came before them. They loosen their boundaries, rearrange their relationships, and give people new things to do with what is inside.
The toy is still sitting on the table. But it can also become a character with behavior, a place someone can enter, or the beginning of another person's creation. Whether we ultimately call that an App, an Agent, a World, or something else matters less than the verbs gathering around it: create, enter, use, share, and Remix.
The content platform of the AI era will not be defined only by what AI can generate. It will be defined by whether people can turn generation into a shared culture: giving an idea behavior, inviting someone else inside, and leaving the work open to what comes next.